Accurate segmentation of fine-scale organs from 3D point clouds poses a substantial challenge in high-throughput plant phenotyping (HTP), where existing methods are hindered by the loss of topological features and the resulting low accuracy. To address this, we introduce CotSkNet, a semantic segmentation network based on structured representation learning. This approach is the first to define plant architecture as a knowledge-based intermediate representation that integrates geometric, topological, and hierarchical relationships, thereby incorporating topological information into the segmentation process. CotSkNet features an innovative Topological Geometric Feature Fusion Extractor and a Two-Way Reinforcement Module to efficiently extract and enhance salient features. Furthermore, the proposed Topological Attention Aggregator enables dynamic focus on key branch points. On a dataset of 403 field-grown cotton plants, our method achieved mean intersection over union values of 91.55%, 92.37%, and 98.21% for the main stem, fruiting branches, and leaves, respectively, far surpassing those of mainstream methods. Moreover, automatically extracted phenotypic parameters, such as plant height and fruiting branch length showed excellent consistency with manual measurements (R2 > 0.91, root mean square error (RMSE) < 0.14). This study confirms that structured representation learning excels at capturing fine-organ phenotypes, providing an innovative analytical pathway for HTP in complex field crops.

Aiming at the problem that Xinmei fruits with different maturity levels in the orchard environment appear to overlap and are heavily shaded by tree trunks, branches and leaves, which leads to the difficulty of recognizing fruit maturity, and the complex orchard environment further exacerbates the difficulty of detection, resulting in the detection of the existence of the inability to correctly identify or the problem of low identification accuracy, this study proposes a lightweight Xinmei maturity detection algorithm model based on the improvement of the YOLOv8s model YOLO-EHS. Firstly, the self-developed PSEFR (Pyramid Scale Enhanced Feature Representation) module is proposed to replace the C2F part in BACKBONE, which has the ability of multiscale feature extraction and enhanced feature fusion, reduces the useless computation, and improves the feature extraction ability of Xinmei fruit. Next, the MSETFPN-Neck (Multi-Scale Enhanced and Efficient Feature Pyramid Network) replaces the original Neck with a lightweight design that integrates multi-scale feature-weighted fusion, segmented efficient convolution, a global heterogeneous kernel selection mechanism, and an efficient upsampling module, enhancing multi-scale feature extraction while reducing model parameters and computation. Finally, a lightweight detection head, LSGED (Lightweight Shared Group Enhanced Detection head), is proposed. It integrates shared convolution, group convolution, and detail-enhanced convolution, effectively improving feature representation and detection accuracy. Experimental results show that the optimized model achieves 93.3% average detection accuracy, with improvements of 1.0% in mAP, 1.8% in precision, and 2.2% in recall over the original model. It also reduces computation by 40.5%, parameters by 45.1%, and model size by 40%. This method effectively addresses the challenge of detecting Xinmei fruits in complex orchard environments, significantly enhancing performance and providing strong support for intelligent picking technology.

The dense tree canopy in the complex orchard environment obstructs wireless positioning signals and generates NLOS interference, which reduces the positioning accuracy of agricultural mobile robots. This study investigates a localization method for agricultural mobile robots based on two UWB tags and an electronic compass. By analyzing the NLOS interference factors and error sources of UWB, a method for NLOS interference suppression and positioning correction employing two UWB tags tightly coupled with heading angle was proposed. The construction of the heading angle L2IB system and its comprehensive process were also introduced as follows. The proposed method constructs candidate localization domains for dual UWB tags based on multilateration and integrates the inter-tag distance and heading-angle constraints within an L2IB framework to suppress NLOS-induced errors and estimate the robot center position. Experiments were performed under four simulated scenarios, namely line-of-sight (LOS), single-anchor occlusion, multi-anchors occlusion, and single-tag occlusion. The proposed method was compared with the centroid and least-squares methods. The results demonstrate that the L2IB method effectively improves localization accuracy under NLOS conditions. Specifically, in the single-tag NLOS interference scenario, the MAE, RMSE, and maximum localization error were 3.7, 4.0, and 6 cm, respectively. These results indicated that the system could meet the positioning needs of most NLOS environments in the orchard. Therefore, the proposed method exhibits feasibility and provides a new alternative for high-precision localization of mobile robots in orchards under NLOS conditions.

IntroductionHigh-throughput field phenotyping (HTFP) holds great potential for elucidating the relationship between genomes and phenotypes. However, obtaining high-quality three-dimensional point cloud data of field populations and achieving single-plant phenotypic analysis remain challenging.MethodsThis study develops an integrated framework for field crop reconstruction based on 3D Gaussian splatting, incorporating a geometry-aware dynamic constraint algorithm to achieve instance segmentation and extract key phenotypic traits of individual plants. Using 3D Gaussian splatting technology, field-scale cotton population modeling is accomplished, generating dense 3D point clouds for regions of interest. Furthermore, the concept of a crop localization domain is proposed, establishing a longitudinal mapping that associates plant positional coordinates with long-term phenotypic attributes. Finally, through a dynamic spatial constraint mechanism, the accuracy and computational efficiency of instance segmentation for crop population point clouds are significantly improved, enabling rapid extraction of individual plant traits such as cotyledon node height, plant height, and leaf area.ResultsThe results demonstrate that PhenotypeAI successfully reconstructed nine cotton populations with PSNR exceeding 30.0 dB. It successfully extracted regions of interest from 403 cotton plants, achieving an average F-score of 91.32% for instance segmentation and an average accuracy of 91.35%. The extracted traits—cotyledon node height, plant height, and leaf area—exhibited strong correlations with manual measurements, with coefficients of determination (R2) of 0.90, 0.91, and 0.91, respectively.DiscussionThe proposed method provides a low-cost solution for high-throughput field phenotypic analysis of field cotton and improves the efficiency of cotton breeding.

To achieve high-precision trajectory tracking for multi-joint robotic manipulators in the presence of model uncertainties, external disturbances, and strong coupling effects, this paper proposes a nonsingular fast terminal sliding mode control (NFTSMC) scheme incorporating an extended Kalman filter-based disturbance observer. First, the Kalman filter is combined with an extended state observer to perform the real-time observation of both internal and external disturbances in the system, accurately estimating system uncertainty and external disturbances. This approach reduces noise interference while significantly improving the correction accuracy of position and tracking errors. Second, an improved nonsingular fast terminal sliding mode controller with an optimized convergence law is introduced to ensure stability during the tracking process, effectively mitigate oscillation phenomena, and accelerate the system’s convergence speed. Finally, the convergence of the proposed method is analyzed by constructing an appropriate Lyapunov function. Simulation and experimental results strongly validate the superior performance of the proposed control strategy, demonstrating that the system can achieve high-precision trajectory tracking under the complex coupled effects of a six-axis robotic manipulator, and exhibits significant advantages in terms of accuracy and robustness.

Accurate and robust recognition of daylily flower buds at the pre-bloom stage is essential for timely harvesting and quality preservation, yet remains highly challenging under natural field conditions due to the buds' slender morphology, diverse orientations, dense distribution, and frequent occlusion by foliage. Existing horizontal-box object detectors struggle to capture the orientation and geometric structure of daylily buds, leading to inaccurate localization and unreliable guidance for automated harvesting. To address these challenges, we propose DDM-YOLO, a lightweight orientation-aware detection model tailored for daylily production environments. The model integrates three key components: (i) a Multi-scale Adaptive Feature Pyramid Network (MAFPN) that enhances the extraction and fusion of multi-dimensional features for densely distributed and occluded slender buds; (ii) a Lightweight Adaptive Direction-aware Head (LADH) that dynamically optimizes angle regression for rotated bounding boxes, improving orientation stability and reducing localization bias; and (iii) an Adaptive Down-sampling module (Adown) that preserves structurally critical spatial cues while reducing model complexity. Experiments conducted on a custom daylily field dataset demonstrate that DDM-YOLO achieves 96.8% precision and 98.1% mAP50, outperforming the baseline YOLOv11n-OBB by 1.3 percentage points in mAP while reducing model parameters by 17.0% to 2.2M. Deployment verification using a PySide5-based visualization prototype demonstrated a total system-level latency of less than 0.2 s, a duration encompassing the cumulative overhead of image input and output, pre-processing, post-processing including non-maximum suppression, and interface rendering. Furthermore, physical deployment on an NVIDIA Jetson AGX Orin embedded platform utilizing TensorRT optimization achieved an impressive inference speed of 114.5 FPS, corresponding to approximately 8.7 ms per frame. This performance confirms that the model meets the stringent real-time requirements for edge computing in mobile agricultural robotics. The model efficiently and accurately performs oriented detection and harvesting pose estimation for daylily buds, providing critical technical support for the visual perception system of harvesting robots.

Agricultural engineering informatics is playing an increasingly important role in enabling intelligent perception, decision-making, and automated operations in modern horticultural production systems. Within this context, accurate visual perception of reproductive structures is essential for agricultural informatization tasks such as flowering-stage monitoring, precision pollination, and information-driven fruit-set management in chili cultivation. However, reliable detection and pose-aware recognition of chili flowers remain challenging because of small target size, dense distribution, foliage occlusion, and illumination variability in natural or semi-controlled environments. To address these challenges, this study proposes a lightweight and robust edge vision framework, termed CFPR-YOLO, for chili flower detection and pose-aware perception under complex agricultural conditions. Built upon an improved YOLOv11n architecture, the proposed framework incorporates EfficientFormerV2 to strengthen global-context feature extraction, a C3k2_EMA module to enhance localization of small and occluded targets, and Poly-Scale Convolution (PSConv) to preserve structural details while reducing computational redundancy. In addition, a lightweight attention mechanism is introduced to improve feature discrimination in cluttered backgrounds. Experimental results on both self-constructed and generalization datasets show that the proposed method achieves a precision of 92.6%, a recall of 86.8%, and an mAP50 of 92.1% with only 7.26 M parameters. The framework also demonstrates strong robustness and generalization across different chili varieties. When deployed on an edge computing platform (NVIDIA Jetson AGX Orin), the model achieves real-time inference at 39.5 FPS. Furthermore, validation experiments under controlled indoor conditions show that the proposed framework can effectively support simulated pollination tasks, achieving a success rate of 90.0% for upwardfacing flowers. These results indicate that CFPR-YOLO provides an effective visual perception solution for agricultural engineering informatics-oriented pollination systems and offers practical potential for precision pollination and intelligent fruit-set management in horticultural production.

To address the challenges of rapid and accurate grape variety identification in natural orchard environments, along with the demand for efficient deployment on mobile devices, we propose in this paper YOLOv11n-GrapeLite, a lightweight model built upon an enhanced YOLOv11n architecture. First, an Efficient Channel Attention (ECA) mechanism is incorporated into the Neck layer. This mechanism adaptively recalibrates feature channel weights to emphasize those relevant to grape variety recognition, suppress background interference, and enhance target feature perception in complex scenes. Second, an adaptive downsampling (ADown) strategy is employed to replace the traditional convolutional downsampling module, reducing computational complexity while preserving critical features. Finally, the original C3k2 module is redesigned as a multi-scale convolution block (MSCB). This block integrates depthwise separable convolutions with multi-scale convolutions, which achieves significant parameter compression and enhances multi-scale feature extraction. Experimental results demonstrate that the proposed model achieves a mean average precision (mAP) of 91.5%, representing a 0.2% improvement over the original YOLOv11n, along with a 0.6% increase in recall. These results indicate outstanding robustness in complex field scenarios. The model’s parameter count was reduced to 1.87 million, computational complexity to 5.0 GFLOPS, and model size to 4.1 MB. These figures represent reductions of 27.8%, 23.1%, and 25.5%, respectively, compared to the original YOLOv11n, demonstrating significant lightweight optimization. Compared to mainstream models such as YOLOv6, YOLOv8n, YOLOv9s, YOLOV12, YOLOv13 and YOLOv26, the proposed model achieves superior performance in parameter count, computational load, and model size, while maintaining competitive detection accuracy. The YOLOv11n-GrapeLite model efficiently adapts to mobile terminal deployment, providing a feasible and efficient technical solution for real-time, precise identification of grape varieties in complex field scenarios.

Precise identification of chili flowers and buds is essential for enabling intelligent robotic pollination, continuous crop monitoring, and early yield prediction in protected horticulture. Reliable visual sensing remains challenging because chili targets are extremely small, densely distributed, and frequently occluded by foliage, while greenhouse environments introduce strong illumination variations and background reflections. These factors often lead to insufficient feature extraction and unstable detection accuracy in existing models, limiting their practical deployment in automated monitoring systems. To address these challenges, this study proposes GCP-YOLO, a lightweight yet high-performance detection framework built upon the YOLOv11n architecture. The model enhances small-target perception through three key improvements. First, a redesigned Generalized Feature Pyramid Network (GFPN) strengthens cross-scale feature interaction, improving the fusion of fine-grained texture cues and deep semantic information. Second, a C2CGA context-guided attention module is introduced to emphasize floral structural features while suppressing background noise caused by reflections and canopy clutter. Third, extended multi-scale detection heads (P2-P6) incorporate broader contextual information to reduce missed detections and false positives in dense planting scenarios. Experimental results on a custom chili flower dataset show that the proposed method achieves 92.8% precision, 83.7% recall, 90.8% mAP50, and 72.7% mAP50-95, improving upon the YOLOv11n baseline by 2.1, 1.3, 3.9, and 6.6 percentage points, respectively. Deployment on an NVIDIA Jetson AGX Orin edge platform demonstrates real-time inference at 97.9 FPS, confirming its suitability for on-device phenotyping. Overall, the proposed approach significantly improves detection robustness under complex greenhouse conditions, providing an effective visual sensing methodology for automated crop monitoring and data-driven yield estimation.

Accurate perception of the environment is paramount in robotic manipulation, driving an increasing reliance on multimodal sensing systems. Vision and tactile sensing emerge as two pivotal modalities; however, prevailing fusion strategies, often based on direct feature concatenation, fail to adequately model the complex spatiotemporal interdependencies between them. This limitation becomes pronounced in dynamic interaction scenarios, where the inability to capture fine-grained visuotactile correlations severely restricts the perception of critical events like object slippage. To address this challenge, we propose a novel multimodal fusion framework based on Tucker decomposition. Our approach introduces three key innovations: independent factor matrices that decouple visual and tactile features into separate latent spaces to mitigate modal confusion; a cross-core tensor fusion strategy leveraging a shared core tensor to explicitly model high-order cross-modal interactions; and an efficient tucker tensor decomposition fusion (TTDF) module, built upon a Transformer architecture, which incorporates hierarchical fusion and dynamic gated attention to adaptively recalibrate modal contributions, improving generalization and robustness. Extensive experiments demonstrate superior performance, achieving state-of-the-art accuracy in robotic slippage detection. The framework enhances visuo-tactile fusion precision, reduces object damage rates in production, increases operational efficiency, and provides crucial technical support for advanced flexible robotic manipulation.

Mechanized harvesting constitutes a pivotal advancement for facilitating the scalability and intelligent evolution of the daylily industry. The core technological challenge lies in the rapid and precise recognition and spatial localization of mature daylily fruits and their connecting stems, which serve as the harvesting points. However, intelligent harvesting in unstructured field environments confronts severe challenges. Mature buds exhibit slender morphologies and random growth postures, resulting in significant inter-target occlusion. Simultaneously, the pedicel, acting as the critical operation point for the robotic end-effector, is minute in size, with texture features highly indistinguishable from the background stems and foliage. Traditional Horizontal Bounding Box (HBB) detection methods fail to accurately characterize the geometric posture of such slender targets and introduce excessive background redundancy, rendering them incapable of providing reliable guidance for precise robotic grasping and cutting. Although existing generic Oriented Bounding Box (OBB) models can generate directional detection boxes, they typically suffer from complex network architectures and high computational overhead, hindering real-time inference on resource-constrained field mobile platforms.To address these issues, this paper proposes a high-performance oriented object detection network named DS-YOLO (Daylily-Stem YOLO). Built upon the state-of-the-art YOLOv11-OBB baseline architecture, the model integrates three targeted optimization strategies. First, a P6 large-scale detection layer is added to the original feature pyramid to expand the effective receptive field of the network, thereby mitigating feature truncation issues for large targets in close-range imagery. Second, a Lightweight Shared Convolutional Decoupled (LSCD) head is designed. This module reconstructs the decoupled head structure using parameter sharing and depthwise separable grouped convolutions, significantly reducing computational costs while ensuring regression precision for mature buds and pedicels. Finally, a C2f-BiLevel Routing Attention (C2BRA) module is proposed to replace the original C2PSA module. Through a dual-level routing mechanism, it realizes dynamic contextual feature filtering, enhancing the sparse perception capabilities of the network regarding minute pedicel features and suppressing background noise. Experimental results on a self-constructed Daylily-OBB dataset demonstrate that, compared to the baseline YOLOv11n-OBB, DS-YOLO achieves improvements of 1.8, 1.4, 1.6, 1.1, and 0.6 %age points in Precision, Recall, F1 score, mAP50, and mAP50-95, respectively. Its comprehensive performance significantly outperforms RT-DETR and other mainstream YOLO algorithms. This study provides a robust technical foundation for high-precision and high-efficiency visual perception and localization in automated daylily harvesting equipment.

At present, fruit picking mainly relies on manual operation. Taking the litchi (litchi chinensis Sonn.)-picking robot as an example, visual perception is often affected by illumination variations, low recognition accuracy, complex maturity judgment, and occlusion, which lead to inaccurate fruit localization. This study aims to establish an embodied perception mechanism based on “perception-reasoning-execution” to enhance the visual perception and decision-making capability of the robot in complex orchard environments. First, a Y-LitchiC instance segmentation method is proposed to achieve high-precision segmentation of litchi clusters. Second, a generative artificial intelligence model is introduced to intelligently assess fruit maturity and occlusion, providing auxiliary support for automatic picking. Based on the auxiliary judgments provided by the generative AI model, two types of dynamic harvesting decisions are formulated for subsequent operations. For unoccluded main fruit-bearing branches, a skeleton thinning algorithm is applied within the segmented region to extract the skeleton line, and the midpoint of the skeleton is used to perform the first type of localization and harvesting decision. In contrast, for main fruit-bearing branches occluded by leaves, threshold-based segmentation combined with maximum connected component extraction is employed to obtain the target region, followed by skeleton thinning, thereby completing the second type of dynamic picking decision. Experimental results show that the Y-LitchiC model improves the mean average precision (mAP) by 1.6% compared with the YOLOv11s-seg model, achieving higher accuracy in litchi cluster segmentation and recognition. The generative artificial intelligence model provides higher-level reasoning and decision-making capabilities for automatic picking. Overall, the proposed embodied perception mechanism and dynamic picking strategies effectively enhance the autonomous perception and decision-making of the picking robot in complex orchard environments, providing a reliable theoretical basis and technical support for accurate fruit localization and precision picking.

Accurate detection of Munage grape clusters and abnormal berries in field scenes is hindered by several challenges: mature berries often exhibit colors similar to those of branches and leaves; cluster-level large targets coexist with medium- and small-scale abnormal berry targets; local abnormal cues within full-berry bounding boxes are easily diluted by responses from normal berry skin and waxy bloom; and shallow-level textures, specular highlights, and adjacent berry boundaries may induce false detections. To address these challenges, this study proposes YEIS, a YOLO11n-based detection method for Munage grape clusters and abnormal berries under color-similar backgrounds. Built upon YOLO11n, the proposed method first introduces EMBSFPN to construct multi-scale candidate features, thereby alleviating the scale-representation discrepancy between cluster-level targets and berry-level abnormal targets. Second, an intra-berry frequency–local evidence decoupling module, IB-FLED, is designed to enhance local abnormal cues within full-berry detection boxes through low-frequency appearance estimation, local residual modeling, and morphology-aware response branches. Finally, a semantic-guided recall compensation module, SGRCM, is developed to constrain P3 detail compensation using P4 semantic information refined by IB-FLED, reducing the interference of shallow-level textures, waxy bloom, and specular highlights in abnormal berry localization. Three random-seed experiments were conducted on a self-built field dataset. The results show that YEIS achieves Precision, Recall, mAP50, mAP75, and mAP50–95 values of 83.21%, 79.97%, 88.34%, 83.27%, and 76.91%, respectively. Compared with YOLO11n, the overall mAP50–95 is improved by 1.71 percentage points, and the mAP50–95 for lesion-like abnormal berries is increased by 1.61 percentage points. For scar-like abnormal berries, the F1-score and mAP50–95 are improved by 3.01 and 3.41 percentage points, respectively. Meanwhile, the number of model parameters is reduced from 2.583 M to 2.149 M, corresponding to a reduction of 16.8%. The proposed method improves the detection and localization of abnormal berries under color-similar backgrounds while reducing the parameter count relative to YOLO11n, providing a front-end visual detection approach for digital monitoring, grape-cluster localization, and abnormal-berry recognition in Munage vineyards.

Achieving an optimal plant architecture is a key objective in cotton breeding for enhancing yield potential, and accurate quantification of the fruit branch angle (FBA) is essential for understanding genotype-phenotype relationships and advancing ideotype breeding. However, in-field FBA measurement remains technically challenging due to severe occlusion, variable illumination, and background interference. To overcome these limitations, we propose a streamlined 3D phenotyping framework that integrates 3D Gaussian Splatting (3DGS) with a novel structural segmentation model, the Linear Point Cloud Reverse Model (LPCRM). The framework decomposes reconstructed cotton point clouds into linear micro-elements using RANSAC, followed by geometric clustering via K-means to identify and separate the main stem and fruit branches. This process operates without topological priors or large annotated datasets. Model fidelity assessment shows that 80% of point pairs between the LPCRM and the original 3DGS reconstruction exhibit Euclidean distances <= 0.5 cm. Phenotypic validation using 268 fruit branches from 25 cultivars demonstrates high measurement accuracy, achieving an R2 of 0.874 and an RMSE of 4.01 degrees for FBA extraction. Plant height estimation also shows strong agreement with manual measurements (R2 = 0.915, RMSE = 3.858). Overall, this study presents a lightweight and robust solution for extracting 3D structural traits of field-grown cotton. The proposed framework reduces data dependency, adapts well to complex field conditions, and offers an efficient approach for high-throughput phenotyping and cotton ideotype breeding.

To address the challenge of pose estimation in intelligent long-staple cotton harvesting caused by dense plant distribution and pronounced spatial heterogeneity of bolls, this study introduces a pose estimation method that integrates growth-relation keypoint constraints with vision-language model reasoning. A YOLO for cotton and stem segmentation (YOLO-CSS) model is developed to achieve fine-grained boll-stem segmentation under complex field conditions. Mathematical representations of the fractal characteristics and structural complexity of cotton plants are established, and growth structure modeling is used to analyze boll-stem spatial relationships, providing structural priors for subsequent orientation estimation. The study developed a cotton boll orientation reasoning method based on vision-language model understanding (VLM-OR), where the model evaluates the reliability of stem-growth keypoint extraction, establishes growth direction priors from these keypoints, and incorporates boll attachment direction for rule-based reasoning, enabling orientation estimation under weak-texture conditions. Furthermore, A boll-stem cooperative localization method is formulated through spatial geometric reasoning, using stem-growth keypoints as spatial anchors to derive the 3D picking pose of bolls and compensate for depth-direction positioning errors of the end effector, thus supporting dynamic alignment between perceptual outputs and execution parameters. Experimental results show that the proposed VLM-OR achieves an orientation success rate of 94.1 % in complex scenarios. Additionally, 74 % of depth keypoint errors remain below 3 mm, the orientation-based grasping success rate reaches 80 %, and more than 65 % of picking attempts succeed within a +/- 20 degrees tolerance range. These findings confirm the method's accuracy and operational adaptability, offering strong methodological support for visual perception in long-staple cotton picking robots.

Inconsistent kernel plumpness in raw walnuts introduces severe quality variations in downstream processed goods, threatening overall product standardization. Conventional mechanical grading methodologies are constrained by low throughput and inherently destructive protocols, necessitating the development of rapid, non-invasive alternatives. This study introduces a novel non-destructive evaluation framework for Juglans regia (cultivars Wen 185 and Xinfeng) by leveraging high-resolution X-ray imaging technology. To execute this, we developed a structurally enhanced FVW-YOLO instance segmentation model to accurately delineate internal morphological boundaries, paired concurrently with a BOSS-PLS quality prediction model designed to quantify mass features based on multi-dimensional textural data. Using extensive destructive shell-breaking validations, we categorized overall kernel plumpness into four discrete empirical grades: plump, moderately plump, slightly shriveled, and shriveled. The integrated dual-model grading method achieved an exceptional overall classification accuracy of 95.6%, which constitutes a statistically significant 3.8% improvement over traditional, single-dimensional area-proportion algorithms. Crucially, this integrated approach minimizes the algorithmic misclassification of boundary-threshold samples. And developed a software for the visualization system of fullness detection.These findings present a highly scalable, automated strategy for the internal quality stratification of in-shell nuts, offering the agricultural and food processing industries a robust analytical tool to enhance raw material utilization, standardize product quality

Abstract The capability to provide an adjustable constant-force output is highly valuable for applications such as micromanipulation and precision assembly. The output force of conventional constant-force mechanisms is difficult to adjust. Their constant-force performance is also highly dependent on structural parameters. To address these limitations, this study proposes an adjustable constant-force mechanism that combines positive and negative stiffness. The mechanism consists of trapezoidal and bistable beams, and its constant-force output can be adjusted by varying the preload displacement between the positive- and negative-stiffness modules. First, analytical models of the trapezoidal and bistable beams were established separately, and the force-displacement relationship of the constant-force mechanism under the zero-preload condition was derived. Subsequently, the key design variables were identified through a sensitivity analysis of the structural parameters. The design was optimized using the non-dominated sorting genetic algorithm II, yielding a set of Pareto-optimal solutions, from which a representative compromise solution was selected. Finally, a prototype was fabricated and experimentally tested. Under the zero-preload condition, optimization increased the constant-force stroke from 0.7 to 2.1 mm and reduced the relative output-force fluctuation within the constant-force region from 10% to 5.5%. Moreover, varying the preload displacement enabled stable adjustment of the constant-force output across different force levels. The experimental results confirm the effectiveness and feasibility of the proposed mechanism and optimization method.

All-day detection and occlusion scenario analysis are critical vision techniques to ensure the continuous operation of tomato-picking robots. Most detection algorithms struggle with varying weather and light. To improve all-day tomato detection accuracy and robustness, this study introduces a lightweight, high-precision method combining deep learning and image processing. At night, HSV color distances between mature tomatoes and backgrounds were calculated under different light, creating an all-day image dataset. An improved model EMSYOLOv8 based on YOLOv8n introducing EfficientViT network, MCA attention mechanism, and SIoU loss function was trained to accurately detect and locate tomatoes under various conditions. Within the detected bounding boxes of mature tomatoes, image processing techniques such as dynamic OTSU, Hough detection with filtering, and connected component analysis were employed to analyze occlusion scenarios and differentiate between occlusion types (leaf-occluded, branch-occluded, and couple-occluded). This AI vision technology for picking robots provides valuable guidance for optimizing picking strategies. Experimental results show that the improved YOLOv8 model achieved 91.9% mAP on the all-day tomato dataset, representing improvements of 1.2, 1.8, 2.5 and 2.8 percentage points compared to YOLOv11, YOLOv10, YOLOv8, and YOLOv5 models, respectively. For occlusion scenario analysis, the overall accuracy of occlusion type determination reached at least 82% under couple-occluded. Feasibility tests in greenhouse environments demonstrated that the proposed method enabled the robot achieve a single-fruit picking success rate of 80% without occlusion under natural light, and 72% under couple-occluded interference; the single-fruit picking success rate reaches 73% without occlusion under nighttime supplemental light, and 63% under couple-occluded interference, which providing robust visual technology support for the development of all-day tomato-picking robots.

Soluble solids content (SSC) and firmness are key quality attributes for assessing the maturity, postharvest quality, and commercial value of Xinmei plum. This study developed a nondestructive analytical approach combining multidimensional correlation spectroscopy with ResNet-34 for SSC and firmness classification, followed by integrated quality grading. Visible/near-infrared (Vis/NIR) diffuse reflectance spectra of 600 Xinmei plum samples were collected over 400–1000nm, and reference SSC and firmness values were measured. One-dimensional spectra were transformed into two-dimensional and three-dimensional correlation spectra using single-quality-gradient perturbation, and synchronous, asynchronous, and integrated spectral representations were compared. Using ResNet-34 as the principal backbone, the 3D-COS synchronous representations achieved validation accuracies of 95.33% for SSC and 91.95% for firmness, respectively, outperforming the corresponding 1D-CNN baselines. Feature visualization using t-SNE and Grad-CAM showed improved extraction and separation of discriminative spectral information related to quality levels. The principal practical innovation was a limiting-factor-based fusion strategy that converted the separate SSC and firmness predictions into a unified commercial grade, achieving an integrated grading accuracy of 95.00% and reducing severe cross-grade misclassification. These results indicate that multidimensional correlation spectroscopy coupled with deep residual learning is effective for nondestructive evaluation and grading of Xinmei plum quality.

In real orchard environments, harvesting robots often operate in dense foliage, narrow inter-row spaces, and under complex, dynamically changing illumination. These conditions limit the robot's ability to obtain wide, continuous viewpoints, with common visual features between sequential frames often obscured by environmental disturbances. This leads to inherently weak-covisibility inputs, characterized by unstable spatial features due to alternating direct sunlight and shadows, small overlapping regions caused by hierarchical occlusions from branches and leaves, and low consistency in morphological and spatial feature distributions across different fruit species. Consequently, robots struggle to build a coherent understanding of orchard scenes and fruit targets, which limits performance in fruit volume estimation, yield assessment, fruit localization, and harvesting path pl anning.To address these challenges, this study analyzes the limitations of existing point cloud registration methods, particularly their reliance on initial poses, feature alignment, and error convergence. We propose a novel point cloud reconstruction framework that integrates iterative semantic cues and geometric priors into a unified optimization mechanism. By jointly constraining geometric priors and semantic segmentation, the proposed method establishes a dynamic iterative optimization process that significantly enhances the stability and accuracy of point cloud registration and surface reconstruction under weak-covisibility conditions. Quantitative comparisons demonstrate that the proposed method achieves a registration RMSE of 1.02 mm under weakcovisibility conditions. Field experiments in real orchard environments further demonstrate robust reconstruction under illumination variations, occlusions, and robotic motion, achieving a fruit reconstruction error of 8.34% and an average reconstruction time of 1.87 s per fruit.This work provides effective support for 3D modeling and autonomous operation of agricultural robots.