Gloss perception, a crucial visual ability, poses computational challenges for human and machine vision when estimating surface glossiness. The Dichoptic Color Difference Threshold (DCDT) is a quantitative metric to assess the minimal discernible color variation in binocular luster perception. In this study, we investigated the relationship between binocular luster and color hue, color difference, and luminance factors by quantifying and measuring DCDTs under binocular equivalent luminance conditions. Additionally, we fitted a quadratic logarithmic function model to describe the association between binocular luster and luminance factors. The experiments and data analysis presented in this article demonstrate that luminance significantly influences binocular luster. These research findings provide robust support for luster reproduction in three-dimensional system design.

Multispectral sensing offers complementary spectral cues that can mitigate the failure modes of single-modal RGB detection under low contrast, illumination changes, and occlusions commonly observed in orchards. Targeting Unmanned Aerial Vehicle (UAV) imagery of walnuts, this paper regards RGB and the red-band (R; 650 +/- 16 nm) as a practical visible-domain pair and proposes a hierarchical two-stream fusion framework precisely aligned with the YOLO feature pyramid. Concretely, we design three cooperative modules-Hierarchical Cross-Modal Dynamic Fusion Mechanism (HCM-DFM) at P3 for fine-grained alignment of geometry and texture across modalities, Cross-Modal Channel-Spatial Fusion (CM-CSF) at P4 for steady recalibration, and Cross-Modal Dynamic Routing Collaboration (CM-DRC) at P5 for long-range, semantics-aware consistency. In support of this study, we build WU-MID, to our knowledge the first RGB-R walnut detection benchmark, consisting of 717 registered image pairs and 63,504 annotated instances, all small objects with dense distribution. On WU-MID, the best mixed configuration - HCM-DFM at P3, CM-CSF at P4, and CM-DRC at P5 - outperforms the two-stream baseline in mAP, AP50, AP75, Precision, and Recall under controlled complexity, and ablation experiments confirm that each scale makes a non-redundant contribution. Cross-domain experiments on FLIR, VEDAI, and DroneVehicle indicate that the proposed scale-wise division of labor generalizes to heterogeneous sensors and scenes. Qualitative visualizations further reveal that the three modules respectively strengthen local alignment, suppress cross-modal noise, and enforce global semantic coherence. The framework is easy to integrate and shows promise for UAV-based orchard applications such as fruit counting, yield estimation, and anomaly monitoring.

Objective The watercolor effect is a classic boundary-induced color filling-in illusion, in which color appears to spread across a region enclosed by chromatic contours. Its formation is thought to rely on the integration of boundary information and surface perception, reflecting fundamental mechanisms of visual processing. While previous studies have extensively examined the roles of contour configuration and chromatic contrast, the contribution of binocular input consistency to the formation of this illusion remains unclear. Understanding whether consistent visual input from both eyes is necessary for the emergence of the watercolor effect is important for clarifying the neural mechanisms underlying binocular integration and perceptual filling-in. Therefore, the present study investigates how dichoptic and binocular viewing conditions influence both the perceptual experience and neural correlates of the watercolor effect. Methods To address this question, this study combines psychophysical experiments with functional near-infrared spectroscopy (fNIRS). Two viewing conditions are designed: dichoptic vision, in which each eye is presented with different color stimuli, and binocular vision, in which both eyes simultaneously view identical stimuli. Fifteen university students aged 23-27 years participate in the study and report their perception of the watercolor effect under each condition, allowing a direct comparison of perceptual outcomes across conditions. Meanwhile, fNIRS is used to record hemodynamic responses in the prefrontal cortex during stimulus presentation. This multimodal approach enables us to examine both behavioral performance and neural activity associated with the illusion under different binocular input conditions. Results and Discussions Behavioral results reveal a clear dissociation between the two conditions. Under binocular vision, participants consistently report a stable and vivid watercolor effect. In contrast, under dichoptic vision, observers generally fail to perceive a coherent or stable filling-in effect, indicating that disrupting interocular consistency significantly impairs the formation of the illusion. These findings provide direct evidence that consistent binocular input is critical for the perceptual manifestation of the watercolor effect. However, the fNIRS results show no significant differences in the average level of prefrontal activation between the two viewing conditions. This suggests that, despite clear perceptual differences, the overall hemodynamic response in the prefrontal cortex does not reliably reflect the presence or absence of the illusion. The dissociation between behavioral and neural results implies that the watercolor effect may be primarily mediated by early visual processing stages involved in binocular integration, rather than higher-level cortical regions. Conclusions In summary, the present study demonstrates that the watercolor effect depends strongly on consistent visual input from both eyes and is disrupted under dichoptic viewing conditions. The absence of significant differences in prefrontal activation further indicates that this region may not serve as a sensitive neural marker for distinguishing perceptual differences related to the illusion. These findings highlight the importance of early binocular integration mechanisms in color filling-in phenomena and contribute to a more comprehensive understanding of the neural basis of visual perception.

UAV-based orchard monitoring can support yield estimation and precision management, yet reliable detection of small and densely distributed fruits remains difficult in complex field scenes. Challenges include limited object pixels, cluttered canopy backgrounds, frequent occlusion, illumination variability, and the need for real-time inference on resource-constrained edge devices. We develop Walnut Lightweight-Pruned YOLO (WLP-YOLO), a task-specific and deployment-oriented detector derived from YOLOv8 for UAV walnut detection under edge-computing constraints. WLP-YOLO combines lightweight feature extraction, efficient multi-scale fusion, hardware-friendly structured channel pruning, and target-device inference to balance small-object detection performance and computational efficiency. On the fixed dataset split, WLP-YOLO increased mAP@0.5 from 0.811 to 0.834 while reducing the model to 2.23 M parameters and 6.8 GFLOPs. Although mAP@0.5:0.95 remained comparable to YOLOv8n (0.323 versus 0.322), small-object AP@0.5 increased from 0.554 to 0.693. Further structured pruning reduced the computational cost to 2.7 GFLOPs, with only a 0.2 percentage-point decrease in mAP@0.5. On the Jetson Xavier NX, the pruned model achieved a per-image latency of 29.1 ms under TensorRT-FP16 inference, corresponding to approximately 34.4 FPS. These results support a practical accuracy–efficiency trade-off for UAV-based walnut monitoring, while transfer to other tasks requires task-specific retraining and cross-domain validation.

Using higher-resolution feature maps in the network is an effective approach for detecting small objects. However, high-resolution feature maps face the challenge of lacking semantic information. This has led previous methods to rely on downsampling feature maps, applying large-kernel convolution layers, and then upsampling the feature maps to obtain semantic information. However, these methods have certain limitations: first, large kernel convolutions in deeper layers typically provide significant global semantic information, but our experiments reveal that such prominent semantic information introduces background smear, which in turn leads to overfitting. Second, deep features often contain substantial redundant information, and the features of small objects are either minimal or have disappeared, which causes a degradation in detection performance when directly relying on deep features. To address these issues, we propose a high-resolution network based on local contextual semantics (HR-SemNet). The network is built on the proposed high-resolution backbone (HRB), which replaces the traditional backbone-FPN architecture by focusing all computational resources of large kernel convolutions on highresolution feature layers to capture clearer features of small objects. Additionally, a local context semantic module (LCSM) is employed to extract semantic information from the background, confining the semantic extraction to a local window to avoid interference from large-scale backgrounds and objects. HRSemNet decouples small object semantics from contextual semantics, with HRB and LCSM independently extracting these features. Extensive experiments and comprehensive evaluations on the VisDrone, AI-TOD, and TinyPerson datasets validate the effectiveness of the method. On the VisDrone dataset, which contains a large number of small objects, HR-SemNet improves the mean average precision (mAP) by 4.6%, reduces the computational cost (GFLOPs) by 49.9%, and decreases the parameter count by 94.9%.

This brief presents SleepNet-XGB, a lightweight hybrid framework for automatic sleep stage classification from single-channel EEG signals. The model decouples representation and decision learning by integrating a CNN-LSTM encoder for multiscale temporal feature extraction with a Bayesian-optimized XGBoost classifier for adaptive decision making. This representation-decision decoupling design effectively balances deep feature expressiveness and classifier robustness under severe class imbalance. Evaluations on three public datasets-Sleep-EDF-20, Sleep-EDF-78, and SHHS-demonstrate superior performance, achieving up to 87.7% accuracy, 0.83 Cohen's kappa, and consistent gains in macro F1-score over recent state-of-the-art methods. With only 0.26 M parameters and fully automated hyperparameter tuning, SleepNet-XGB provides high generalization and computational efficiency, making it well suited for real-time and low-power wearable biomedical systems.

Conventional “sample-destructive—offline” methods for determining the Mooney viscosity (ML) of natural rubber (NR) are time-consuming, costly, and unsuitable for on-site process control. In this study, we propose an end-to-end, non-destructive framework that directly predicts ML from raw NR sheets using visible–near-infrared hyperspectral imaging (Vis–NIR HSI, 420–870 nm, 150 sampled bands at 2 nm intervals) coupled with a deep regression network. A total of 540 NR specimens were imaged under strict darkroom conditions. We developed a one-dimensional convolutional neural network augmented with a Squeeze-and-Excitation (SE) spectral attention module—termed SE-Attention-1D-CNN—to extract multi-scale spectro-chemical features while adaptively reweighting wavelength channels associated with viscosity-related moieties. The model was trained using an L2-regularized mean squared error (MSE) loss, the Adam optimizer, four-fold cross-validation, and early stopping to ensure generalization.The proposed network achieved an average correlation coefficient(R) of 0.873, coefficient of determination (R2) of 0.762, root mean square error (RMSE) of 4.037, mean absolute error (MAE) of 3.364, and residual predictive deviation (RPD) of 2.003. Ablation experiments further demonstrated the effectiveness of the SE module. Specifically, after removing the SE module, the CNN model showed inferior performance, with R and R2 decreasing to 0.827 and 0.684, respectively, RMSE and MAE increasing to 4.645 and 3.775, respectively, and RPD dropping to 1.761. These results highlight the crucial role of attention-driven spectral weighting in improving predictive accuracy and model robustness. By embedding the SE mechanism within a 1D CNN, this study enables high-precision, non-destructive ML prediction and provides a novel approach for the intelligent grading of natural rubber in industrial settings.

In the era of information overload, understanding the brain's adaptive responses to prolonged cognitive tasks is critical. This study investigates the neural compensatory mechanisms that sustain cognitive performance under mental fatigue, offering insights into dynamic resource allocation and practical applications in high-demand settings. Twenty healthy participants performed a Stroop-based cognitive interference task while prefrontal hemodynamic activity was monitored using functional near-infrared spectroscopy (fNIRS). Subjective fatigue was assessed via the Multidimensional Fatigue Inventory (MFI-20), NASA Task Load Index (NASA-TLX), and Visual Analogue Scale (VAS). Behavioral performance (reaction time and accuracy) was recorded simultaneously. Neural activation was analyzed using a Generalized Linear Model (GLM), and functional connectivity alongside network topology metrics (global efficiency, clustering coefficient) were evaluated. Results show that subjective fatigue increased significantly post-task (MFI-20, p < 0.05), with progressive rise in VAS scores. Behaviorally, reaction times decreased while accuracy remained stable, indicating a speed-accuracy trade-off. fNIRS revealed marked activation changes in specific prefrontal regions (e.g., CH1, CH7), with overall activation shifting from positive to negative. This pattern may reflect time-dependent modulation of task-evoked activation and could be associated with multiple factors, including fatigue-related changes in engagement, habituation effects, or resource-related processes. In addition, fatigue accumulation was accompanied by increased functional connectivity between the frontal eye fields (FEF) and dorsolateral prefrontal cortex (DLPFC) (F = 4.61, p = 0.008), as well as between the frontopolar area (FPA) and DLPFC (F = 3.74, p = 0.020). Global efficiency (F = 0.169, p = 0.022) and clustering coefficient (F = 0.177, p = 0.008) also showed significant increases across task progression.Together, these findings may indicate time-dependent modulation of prefrontal network organization during prolonged cognitive interference tasks. Rather than reflecting a single mechanism, these changes could be associated with dynamic adjustments in functional coordination under sustained task demands. The present findings may provide preliminary neurophysiological evidence relevant to neuroergonomics, brain-computer interfaces, and cognitive workload management.

Infrared–visible image fusion aims to highlight thermal targets while preserving fine visible textures. Deep-unrolled low-rank representation learning offers an interpretable fusion paradigm, but it is often constrained by limited context modeling and vulnerable to small residual misalignments when operating at full resolution, which can cause ghosting and edge splitting. To address these issues, we propose LASP-LRRNet, a linear atrous spatial pyramid (L-ASPP)-enhanced unrolled LRR fusion network with Efficient Multi-scale Attention (EMA) and an Adaptive Fine Alignment (AFA) module. L-ASPP is tailored to the LRRNet-style unrolled iterations by replacing the original convolutional update with a lightweight parallel dilated-convolution aggregation, enlarging the effective receptive field and capturing multi-scale context. EMA further strengthens informative features via efficient joint channel–spatial reweighting, and AFA corrects few-pixel shifts at the feature level to suppress alignment-related artifacts. Extensive experiments on public benchmarks demonstrate consistent improvements in texture clarity, target saliency, and artifact suppression compared with existing methods.

Objective Cognitive conflict, a core component of executive control, has historically been investigated through the framework of externally driven interference, most notably the classic Stroop paradigm. In such tasks, conflict emerges passively when incongruent perceptual attributes-for instance, the word "RED" presented in green ink-compete for attention and processing resources. Recent evidence, however, indicates that the human brain can actively generate cognitive conflict through endogenous mental imagery, even in the absence of conflicting external stimuli. This capacity to internally simulate conflict provides a unique perspective for examining top-down cognitive control mechanisms. Despite its theoretical importance, direct neurophysiological evidence remains limited concerning how different mental imagery strategies shape neural activation patterns within the prefrontal cortex (PFC). A central unresolved question is whether the PFC relies on a static representational template, such as a fixed color schema, or dynamically reconfigures its representational geometry according to task demands. For example, does imagining the semantic meaning of a word engage the same PFC substrates as imagining its perceptual color? Moreover, the functional dissociation between the left and right dorsolateral prefrontal cortex (dlPFC) and the contribution of the medial prefrontal cortex (mPFC) in forming such endogenous representations, remain poorly understood. To address these gaps, this study examined the neural mechanisms that underlie endogenous color-imagery conflict. We hypothesize that distinct imagery strategies will evoke differential and lateralized activation patterns in the dlPFC, reflecting functional specialization between semantic decoupling and the integration of multidimensional features. Methods We developed a novel Stroop color-imagery task with high-density functional near-infrared spectroscopy (fNIRS) to test our hypotheses. Fifteen healthy right-handed adults (mean age: 24 +/- 2 years) participated. The stimuli were calibrated in the CIELAB color space using opponent-color pairs (red-green and blue-yellow) with controlled luminance. In the task, participants adopted two imagery strategies upon viewing incongruent Stroop words: 1) word-meaning imagery (imagining the color named by the word, ignoring the ink color), and 2) ink-color imagery (imagining the physical ink color, ignoring the word meaning). Neural activity was recorded using a NirSmart-3000K fNIRS system covering the bilateral dlPFC and the mPFC, registered to Montreal Neurological Institute (MNI) space. We employed representational similarity analysis (RSA) to examine the geometry of neural representations, rather than activation intensity alone. RSA correlates activation patterns across conditions, revealing how information is organized. Two key contrasts were analyzed: 1) the same stimulus with different imagined colors (testing feature integration), and 2) different stimuli with the same imagined color (testing semantic-perceptual decoupling). Preprocessing included wavelet-based motion correction and 0.01-0.1 Hz bandpass filtering. Results and Discussions Representational similarity analysis (RSA) revealed that the prefrontal cortex (PFC) dynamically constructs color representations according to task goals, demonstrating clear hemispheric specialization. In the "same stimulus, different imagined colors" condition, which required switching between semantic and perceptual features of the same input, the right dlPFC showed high representational similarity. This suggests that the right dlPFC integrates multidimensional features-binding static visual input with flexible semantic labels to form coherent mental images, consistent with its role in holistic and cross-modal processing. In contrast, the "different stimuli, same imagined color" condition, which required extracting the same color from different incongruent stimuli, elicited systematic representational differentiation in the left dlPFC. Activation patterns were negatively correlated across stimuli, indicating that the left dlPFC actively decouples semantic meaning from perceptual interference, aligning with its proposed role in analytical control and response inhibition. Notably, the medial PFC (mPFC) did not exhibit stable color-specific representations in either condition, suggesting it is not directly involved in encoding endogenous color imagery, despite its engagement in higher-order cognitive and self-referential processes. These findings challenge the view of the dlPFC as a unitary inhibitory hub. Instead, we propose a dual-pathway regulatory model: the left dlPFC supports analytical decoupling (filtering irrelevant information), while the right dlPFC facilitates synthetic integration (binding relevant features). Their cooperation enables flexible resolution of endogenously generated conflict. The consistency of these effects across color pairs (red-green and blue-yellow) further indicates that this lateralization reflects a general mechanism of cognitive control, rather than being tied to specific visual pathways. Conclusions This study presents novel neurophysiological evidence that endogenous cognitive conflict in the Stroop task is managed by a specialized and lateralized prefrontal network. By integrating high-density fNIRS with representational similarity analysis, we show that the right dlPFC is critically involved in integrating multidimensional features to construct mental images under stable stimulus conditions, whereas the left dlPFC is essential for decoupling semantic and perceptual information to extract target concepts across varying inputs. Contrary to some hypotheses, the medial PFC does not contribute directly to generating specific imagined color content. This study extends the Stroop paradigm to mental imagery, revealing that the prefrontal cortex functions as a flexible builder with distinct left-right dlPFC roles, and validates fNIRS-RSA as a powerful tool for decoding top-down cognitive control.

Infestation by the Red Turpentine Beetle is one of the key factors threatening the ecological stability of pine forests, while traditional manual inspection methods are inefficient and easily influenced by subjective factors. UAV-based detection methods can significantly improve monitoring efficiency; however, small target scales, blurred details, and complex backgrounds in UAV imagery pose great challenges to detection algorithms. To address these issues, This paper proposes TLK-YOLO, a UAV-based method for detecting pine trees infested by the Red Turpentine Beetle, which integrates a Transformer mechanism with a large-kernel selection strategy. A Local Window Cross-Attention (LWCA) upsampling module is introduced to enhance small-object detail modeling by fusing high- and low-resolution information within local windows. A Dynamic Combined Large Selective Kernel (DCLSK) module adaptively adjusts the receptive field to balance global context and local details, improving small-object feature discriminability and robustness. Furthermore, a redesigned network architecture, TLK-NA, mitigates redundancy in shallow-to-deep feature propagation by reversing the information flow from deep to shallow layers, achieving a better trade-off between detection accuracy and computational efficiency. Experimental results demonstrate that TLK-YOLO exhibits remarkable performance in detecting Red Turpentine Beetle infestations from UAV imagery, achieving an mAP50 of 89.3%. Compared with the baseline model, TLK-YOLO attains comprehensive improvements in both detection precision and recall, with precision increases by 4.4%, recall precision by 9.6%, while mAP50 and mAP50-95 improve by 7.9% and 8.3%, respectively, while maintaining relatively low computational overhead. This network provides an efficient and reliable technical foundation for UAV-based forestry pest and disease monitoring.

Accurate traffic forecasting is a challenging task due to the complex spatiotemporal dependencies and nonstationary dynamics present in irregular road networks. In this work, we introduce ESTGFormer, a spatiotemporal graph Transformer for multi-horizon forecasting. The model integrates multi-head temporal self-attention with a serial spatial pathway. This pathway first applies global spatial self-attention and is followed by learnable graph convolution, which effectively separates non-local context encoding from topology-aware diffusion. To capture latent, time-varying dependencies, ESTGFormer employs lightweight embeddings that combine periodic priors, such as time-of-day and day-of-week, with node identity and adaptive components. To improve robustness and temporal consistency across time horizons, we propose StructureAwareLoss, which augments Huber loss with a structure-aware regularizer. This regularizer aligns prediction errors over different horizons, reducing training fluctuations. Our experiments on five benchmark datasets (METR-LA, PEMS-BAY, PEMS04, PEMS07, PEMS08) show that ESTGFormer achieves state-of-the-art accuracy while maintaining competitive computational efficiency. Ablation and sensitivity studies confirm the complementary roles of temporal attention, serial spatial modeling, learnable embeddings, and the proposed loss function, making the model capable of providing stable and generalizable multi-step forecasts for intelligent transportation applications.

Accurate walnut yield prediction is crucial for the development of the walnut industry. Traditional manual counting methods are limited by labor and time costs, leading to inaccurate walnut quantity assessments. In this paper, we propose a walnut detection method based on UAV (UAV means Unmanned Aerial Vehicle) remote sensing imagery to improve the walnut yield prediction accuracy. Based on the YOLOv11 network, we propose several improvements to enhance the multi-scale object detection capability while achieving a more lightweight model structure. Specifically, we reconstruct the feature fusion network with a hierarchical scale-based feature pyramid structure and implement lightweight improvements to the feature extraction component. These modifications result in the RSWD-YOLO network (RSWD means remote sensing walnut detection; YOLO means ‘You Only Look Once’, and it is the specific abbreviation used for a series of object detection algorithms), which is specifically designed for walnut detection. Furthermore, to optimize the detection performance under hardware resource constraints, we apply knowledge distillation to RSWD-YOLO, thereby further improving the detection accuracy. Through model deployment and testing on small edge devices, we demonstrate the feasibility of our proposed method. The detection algorithm achieves 86.1% mean Average Precision on the walnut dataset while maintaining operational functionality on small edge devices. The experimental results demonstrate that our proposed UAV remote sensing-based walnut detection method has a significant practical application value and can provide valuable insights for future research in related fields.

Walnut detection in mountainous and hilly regions often faces significant challenges due to obstructions, which adversely affect model performance. To address this issue, we collected a dataset comprising 2379 walnut images from these regions, with detailed annotations for both obstructed and non-obstructed walnuts. Based on this dataset, we propose OW-YOLO, a lightweight object detection model specifically designed for detecting small, obstructed walnuts. The model’s backbone was restructured with the integration of the DWR-DRB (Dilated Weighted Residual-Dilated Residual Block) module. To enhance efficiency and multi-scale feature fusion, we incorporated the HSFPN (High-Level Screening Feature Pyramid Network) and redesigned the detection head by replacing the original head with the more efficient LADH detection head while removing the head processing 32 × 32 feature maps. These improvements effectively reduced model complexity and significantly enhanced detection accuracy for obstructed walnuts. Experiments were conducted using the PyTorch framework on an NVIDIA GeForce RTX 4060 Ti GPU. The results demonstrate that OW-YOLO outperforms other models, achieving an mAP@0.5 (mean average precision) of 83.6%, mAP@[0.5:0.95] of 53.7%, and an F1 score of 77.9%. Additionally, the model’s parameter count decreased by 49.2%, weight file size was reduced by 48.1%, and computational load dropped by 37.3%, effectively mitigating the impact of obstruction on detection accuracy. These findings provide robust support for the future development of walnut agriculture and lay a solid foundation for the broader adoption of intelligent agriculture.

The presence of non-tobacco-related materials can significantly compromise the quality of tobacco. To accurately detect non-tobacco-related materials, this study introduces a lightweight and real-time detection model derived from the YOLOv11 framework, named LRNTRM-YOLO. Initially, due to the sub-optimal accuracy in detecting diminutive non-tobacco-related materials, the model was augmented by incorporating an additional layer dedicated to enhancing the detection of small targets, thereby improving the overall accuracy. Furthermore, an attention mechanism was incorporated into the backbone network to focus on the features of the detection targets, thereby improving the detection efficacy of the model. Simultaneously, for the introduction of the SIoU loss function, the angular vector between the bounding box regressions was utilized to define the loss function, thus improving the training efficiency of the model. Following these enhancements, a channel pruning technique was employed to streamline the network, which not only reduced the parameter count but also expedited the inference process, yielding a more compact model for non-tobacco-related material detection. The experimental results on the NTRM dataset indicate that the LRNTRM-YOLO model achieved a mean average precision (mAP) of 92.9%, surpassing the baseline model by a margin of 4.8%. Additionally, there was a 68.3% reduction in the parameters and a 15.9% decrease in floating-point operations compared to the baseline model. Comparative analysis with prominent models confirmed the superiority of the proposed model in terms of its lightweight architecture, high accuracy, and real-time capabilities, thereby offering an innovative and practical solution for detecting non-tobacco-related materials in the future.

BACKGROUND:In the process of tobacco sorting, removing non-tobacco related materials (NTRMs) is crucial for the quality of tobacco products. Because of the small size of NTRMs and the abundance and stacking of tobacco leaves, detection of NTRMs is still difficult. RESULTS:Based on YOLOv8s (You Only Look Once, version 8, small), the present study proposed an efficient YSFER-Tobacco (YOLOv8s-SPDConv-FasterNet-EMA-RTDETRDecoder-Tobacco) model for detection of NTRMs. We replaced some down sampling convolutions in the backbone with SPDConv modules and reconstructed the C2f module using FasterNet and EMA to reduce redundant convolution operations and improve feature extraction capabilities. Finally, the RTDETRDecoder from RT-DETR was employed to improve the head component, resulting in more efficient end-to-end target identification. Experimental results demonstrate that YSFER-Tobacco achieved good model performance, with F1, mAP50, recall and precision reaching 96.1%, 97.2%, 95.7% and 96.5%, respectively, compared to YOLOv8s, which increased by 2.5%, 0.9%, 3.2% and 1.7%. YSFER-Tobacco also outperformed other classical object detection models for detection of NTRMs in tobacco sorting process. CONCLUSION:Our study demonstrates the effectiveness and superiority of YSFER-Tobacco, providing theoretical support for assessing the quality of tobacco sorting, and has promising application prospects. In addition, we replicated the tobacco sorting environment and created the first dataset, Tobacco-2619, containing 2619 clear images with NTRMs. The dataset and code are available online (https://github.com/Ikaros-sc/Tobacco). © 2025 Society of Chemical Industry.

Most existing small object detection methods rely on residual blocks to process deep feature maps. However, these residual blocks, composed of multiple large-kernel convolution layers, incur high computational costs and contain redundant information, which makes it difficult to improve detection performance for small objects. To address this, we designed an improved feature pyramid network called L Feature Pyramid Network (L-FPN), which optimizes the allocation of computational resources for small object detection by reconstructing the original FPN structure. Based on L-FPN, we further proposed a small object detector named BPD-YOLO. We introduce a Dual-phase Asymptotic Feature Fusion mechanism (DAFF), where the shallow and deep semantic features extracted from the backbone network are initially fused in parallel to mitigate the semantic gap. Subsequently, the intermediate semantic layers are progressively integrated, enabling effective fusion of both shallow and deep feature representations. Additionally, we designed the Deep Spatial Pyramid Fusion module (DSPF), which generates multi-scale feature representations as an alternative to conventional residual block stacking, thereby reducing computational overhead. In the shallow feature extraction stage, DSPF focuses on semantic integration and enhances the extraction of small object features. This strategy, which adaptively selects different modules based on the resolution of the feature maps, is referred to as the Decoupled feature Extraction-semantic Integration mechanism (DEI). Finally, we conducted extensive experiments and thorough evaluations on both the VisDrone and TinyPerson datasets. The results demonstrate that, on the VisDrone dataset, compared to the baseline model YOLOv8n + p2, our BPD-YOLO model with L-FPN achieves a 2.8% improvement in mAP50 and a 1.4% increase in mAP50-95. On the TinyPerson dataset, BPD-YOLO further demonstrates its superiority in high-resolution feature extraction, effectively enhancing detection accuracy while significantly reducing computational costs.

IntroductionUnderstanding how the brain processes color information from both the left and right eyes is a significant topic in neuroscience. Binocular color fusion and rivalry, which involve advanced cognitive functions in the prefrontal cortex (PFC), provide a unique perspective for exploring brain activity.MethodsThis study used functional near-infrared spectroscopy (fNIRS) to examine PFC activity during binocular color fusion and rivalry conditions. The study included two fNIRS experiments: Experiment 1 employed long-duration (90 s) stimulation to assess brain functional connectivity, while Experiment 2 used short-duration (10 s) repeated stimulation (eight trials), analyzed with a generalized linear model to evaluate brain activation levels. Statistical tests were then conducted to compare the differences in brain functional connectivity strength and activation levels.ResultsThe results indicated that functional connectivity strength was significantly higher during the color fusion condition than the color rivalry condition, and the color rivalry condition was stronger than the Mid-Gray field condition. Additionally, brain activation levels during binocular color fusion were significantly greater, with significant differences concentrated in channel (CH) 12, CH13, and CH14. CH12 is located in the dorsolateral prefrontal cortex, while CH13 and CH14 are in the frontal eye fields, areas associated with higher cognitive functions and visual attention.DiscussionThese findings suggest that binocular color fusion requires stronger brain integration and higher brain activation levels. Overall, this study demonstrates that color fusion is more cognitively challenging than color rivalry, engaging more attention and executive functions. These results provide theoretical support for the development of color-based brain-computer interfaces and offer new insights into future research on the brain's color-visual information processing mechanisms.

Enhancing non-tobacco related materials control and improving the purity of tobacco leaves have emerged as pivotal quality indicators for raw material processing in both domestic and foreign industrial enterprises. In order to accurately detect non-tobacco related materials, this paper introduces an enhanced variant of the YOLOv8(You Only Look Once version 8) model, termed NTRM-YOLO. NTRM-YOLO use deep learning methods to detect non-tobacco related materials. The attention mechanism module is integrated into the backbone network of NTRM-YOLO, aimed at enhancing the delineation of non-tobacco related materials features, thereby bolstering the detection efficacy of the model. In order to reduce the number of model parameters, this paper integrates GhostConv(Ghost Convolution) module within the neck network, alongside the design integration of a GhostConv-C2F module. This strategic substitution serves to diminish the model’s parameters while concurrently enhancing its capacity for feature expression. Within the Head network, capitalizing fully on the merits of multiple attention mechanisms, Dyhead(Dynamic Head) is introduced with the aim of markedly enhancing the detection accuracy of the model. This study also optimized the loss function by using the vector angle. Moreover, this paper uses industrial camera sensors to collect images containing non-tobacco related materials and constructed of an NTRM dataset after preprocessing. Subsequently, a meticulously series of experiments was conducted on the NTRM dataset to showcase the efficacy of NTRM-YOLO model in applications pertaining to non-tobacco related materials detection. The experimental findings reveal that in contrast to the baseline model, NTRM-YOLO attained a detection performance of 95.6%, marking a notable improvement of 2% over the baseline model. Additionally, it exhibited a parameters of 10.0 MB, reflecting a 10% reduction compared to the baseline model. These experiments furnishes a theoretical foundation and technical substantiation for the subsequent advancement of more refined industrial impurity removal instruments and equipment.

Images captured from a drone's perspective are significantly impacted in terms of target detection algorithm performance due to the notable differences in target scales and the presence of numerous small target objects lacking detailed information. This paper proposes a Remote Sensing Small Target Detector (CF-YOLO) based on the YOLOv11 model to address the challenges of small target detection. Firstly, addressing the issue of small target information loss that may arise from hierarchical convolutional structures, we conduct in-depth research on the Path Aggregation Network (PAN) and innovatively propose a Cross-Scale Feature Pyramid Network (CS-FPN). Secondly, to overcome the problems of positional information deviation and feature redundancy during multi-scale feature fusion, we design a Feature Recalibration Module (FRM) and a Sandwich Fusion Module. We advocate for initial feature fusion through the FRM module, followed by feature enhancement using the Sandwich module. Finally, we optimize and reconstruct the model using the RFAConv module and LSDECD detection head. Experiments show that on the public VisDrone dataset, TinyPerson dataset, and HIT-UAV dataset, CF-YOLO improves the mAP50 by 12.7%, 10.1%, and 3.5%, respectively, compared to the baseline model. Compared to other methods, CF-YOLO demonstrates superior performance.