Estimating the mass of Oudemansiella raphanipies quickly and accurately is indispensable in optimizing post-harvest packaging processes. Traditional methods typically involve manual grading followed by weighing with a balance, which is inefficient and labor-intensive. To address the challenges encountered in actual production scenarios, in this work, we developed a novel pipeline for estimating the mass of multiple Oudemansiella raphanipies. To achieve this goal, an enhanced deep learning (DL) algorithm for instance segmentation and a machine learning (ML) model for mass prediction were introduced. On one hand, to segment multiple samples in the same image, a novel instance segmentation network named FinePoint-ORSeg was applied to obtain the finer edges of samples, by integrating an edge attention module to improve the fineness of the edges. On the other hand, for individual samples, a novel cap–stem segmentation approach was applied and 18 phenotypic parameters were obtained. Furthermore, principal component analysis (PCA) was utilized to reduce the redundancy among features. Combining the two aspects mentioned above, the mass was computed by an exponential GPR model with seven principal components. In terms of segmentation performance, our model outperforms the original Mask R-CNN; the AP, AP50, AP75, and APs are improved by 2%, 0.7%, 1.9%, and 0.3%, respectively. Additionally, our model outperforms other networks such as YOLACT, SOLOV2, and Mask R-CNN with Swin. As for mass estimation, the results show that the average coefficient of variation (CV) of a single sample mass in different attitudes is 6.81%. Moreover, the average mean absolute percentage error (MAPE) for multiple samples is 8.53%. Overall, the experimental results indicate that the proposed method is time-saving, non-destructive, and accurate. This can provide a reference for research on post-harvest packaging technology for Oudemansiella raphanipies.

Crop ideotype breeding aims to design plant architectures that enhance yield and resource use efficiency. Accelerating this process demands a framework for assembling accurate three-dimensional (3D) architecture. This Perspective synthesizes advances in technologies and methodologies in 3D architectural phenotyping. Rapid progress has enabled translating these advances into tangible gains in breeding efficiency. To this end, we propose integrating functional-structural plant models as an overarching framework that optimizes plant architecture combinations, shifting breeding from experience-driven to predictive ideotype design. Convergence of 3D phenotyping, plant modeling and artificial intelligence holds transformative potential to accelerate breeding cycles, enhancing productivity, sustainability, and food security.

Accurate and non-destructive volume estimation of agricultural products is essential for precision agriculture, yet remains challenging when transitioning from controlled laboratory conditions to complex orchard environments. Although 2D image-based volume estimation methods provide a cost-effective and scalable solution, existing studies are fragmented and lack a unified perspective on their real-world applicability. This review presents a systematic synthesis of 2D image-based volume estimation methods, explicitly framed through the laboratory-to-orchard transition. We categorized existing volume estimation approaches according to the sensing modality into monocular RGB-based approaches and depth-assisted methods, and further reviewed them based on the image processing methods. A key finding is that high-precision geometric estimation can be achieved in laboratory environments, whereas deep learning and RGB-D fusion have driven a shift from conventional geometric modeling toward data-driven and hybrid learning frameworks in orchard settings. However, 2D image-based volume estimation remains fundamentally limited by scale ambiguity, severe occlusion, and sensitivity to illumination and background variability in real orchard environment. Overall, this review provides a unified perspective for understanding volume estimation methodology across environments and offers guidance for developing robust, scalable, and field-deployable volume estimation systems for real-world agricultural applications.

Crack detection on the surface of nuclear cladding coatings is critical for ensuring the safe operation of nuclear power plants. However, due to the imbalance between crack and background pixels, complex crack morphology, numerous interfering factors, and the subtle features of fine cracks in nuclear cladding coating surface images, the detection performance of existing methods remains unsatisfactory. To address these issues, this paper proposes a novel crack detection model for nuclear cladding coatings surfaces, named CrackCTFuse. This model effectively captures both local detailed features and global context in crack images. Additionally, a crack local feature enhancement module is designed to supplement and enhance the edge details information of cracks, and a crack feature fusion module is proposed to facilitate the effective integration of local and global features. Moreover, a multi-scale convolutional attention module based on channel segmentation is developed to aggregate multi-scale contextual information, enhance skip connections, and improve the model’s ability to perceive and represent crack features at various scales. Experiments conducted on the constructed nuclear cladding coating surfaces crack dataset demonstrate the effectiveness and accuracy of the CrackCTFuse model, achieving a MIoU of 92.70% and an F1-score of 92.54%.

High-speed railway traction substations are critical facilities in high-speed rail systems, responsible for providing stable electric power to high-speed trains through the power supply system. However, anomalies endangering the traction substation occur from time to time. Only by detecting anomalies promptly, timely and targeted countermeasures can be adopted. Therefore, the timely detection of anomalies is essential for the safe operation of traction substations. Existing methods can only detect known anomalies; however, anomaly types are diverse, and abnormal data is difficult to collect, resulting in high uncertainty of anomalies. To address the issue of anomaly uncertainty in visual anomaly monitoring of traction substations, this article proposes an anomaly detection method based on out-of-distribution (OOD) detection through joint entropy-energy optimization. Specifically, we construct an anomaly scoring network (AS-Net) to jointly model semantic, entropy, and energy feature spaces. This allows for a more effective characterization of data distributions and differentiation between in-distribution normal data and OOD anomalies. We further construct the hybrid feature anomaly detection network (HFAD-Net) to fuse information from different perspectives of the AS-Net and detect anomalies. Experimental results on the traction substation dataset demonstrate that the proposed method can improve anomaly detection performance, with the AP increased by 30.81% and FPR95 reduced by 27.69% on average compared to other training methods. The final anomaly detection AP reaches 61.67%, surpassing multiple comparison methods. We also validated the generalizability and effectiveness of the proposed method on three benchmark datasets. Overall, this article provides an efficient and accurate anomaly monitoring method for traction substations.

With the development of computer vision techniques, significant breakthroughs have been achieved in closed-set visual recognition tasks. However, in real-world recognition or classification scenarios, it is often challenging to exhaustively collect training examples for all classes due to various constraints. A more realistic scenario is Visual Open-Set Recognition (OSR), where incomplete knowledge exists during model training, and unknown classes may be encountered during testing. This requires a classifier that can accurately categorize known classes and efficiently handle unknown classes. In this paper, we systematically track and summarize the latest research on visual open-set recognition, providing a comprehensive classification and review of current OSR methods, including DNN-based OSR methods and visual language model guided OSR methods. We then present representative approaches for OSR extension tasks and expansion tasks in open-set environments. Subsequently, we analyze and compare the performance of typical and state-of-the-art OSR methods across different datasets. Finally, we discuss some of the remaining challenges and future research directions in the field of visual open-set recognition.

The pantograph-catenary system (PCS) is a key equipment for electric trains to obtain electrical energy from the traction power supply system. As an abnormal phenomenon in the PCS, pantograph-catenary arcing (PCA) directly affects the current collection quality and operational safety of electric trains. Therefore, it is very important to achieve precise detection of PCA. At present, aiming at the difficulties in detecting small arcing and the poor performance of arcing detection in complex environments in the visual detection task of PCA, a semantic segmentation model arcing segmentation (ArcSE) based on feature enhancement is proposed. This model designs a PCA segmentation model that includes semantic feature branches (SFBs), detail feature branches (DFBs), and feature enhancement mechanisms (FEMs). To address the difficulty of detecting small arcing, a dual-branch structure is designed, which utilizes the semantic information extracted from the SFB to adjust the detail feature map in the DFB, filters out interfering features, and retains small arcing features. Aiming at the difficulty of identifying arcing targets in complex scenes, an FEM is designed, merging the arcing features with the detailed features at different scales through a multiscale features fusion strategy. At the same time, based on the learnable visual center module, the difference between arcing features and background features is further strengthened, effectively improving the robustness of the model. Experiments were conducted on the constructed dataset to validate the effectiveness of ArcSE, with a segmentation accuracy of 89.70% and an inference speed of 14.91 ms.

In data-driven decision systems, boundary patterns are instrumental for delineating distinct categories. However, the scarcity of representative samples that encapsulate the class data distribution often results in a dearth of critical boundary information, impeding the comprehension of data structures and the extraction of pivotal details. To surmount the challenge of predicting boundary patterns amidst insufficient data, this paper proposes a class boundary prediction method by self-representation based pattern selection and geometric-constrained diffusion generation (BP-PSG). It leverages the negative components of samples based on self-representation to distinguish between different edge patterns. A novel conditioning mechanism is proposed for diffusion model that constrains data generation along geometrically meaningful directions. The aggregated data is then subjected to a cycle of evaluation and generation to obtain more potential representative samples that reflect the class data distribution. The ultimate edge patterns derived from this comprehensive dataset are deemed as class boundary. The effectiveness of our method has been validated on both high-dimensional and low-dimensional data, including synthetic data, benchmark datasets, and the RSRAC dataset. Notably, BP-PSG enhances anomaly detection accuracy by 5.44 % to 27.38 % on the benchmark datasets. On the RSRAC dataset, it improves individual emitter identification accuracy by an average of 3.95 % and unknown-class detection accuracy by an average of 11.35 %.

Accurate detection and segmentation of traction substation equipment are prerequisites for achieving automated fault detection in the equipment. However, the complex scenes, and similarities among different equipment in the substation pose significant challenges for the detection and segmentation tasks. Meanwhile, existing research methods are specifically designed for either object detection or instance segmentation, making it difficult for them to maintain excellent performance on both tasks. To achieve a unified model for equipment detection and segmentation with improved accuracy, we propose a unified model named Cascade Query Embedding Mask DINO (CQE-Mask DINO) for traction substation equipment detection and segmentation based on Mask DINO. Firstly, we design a novel query embedding interaction module (QEIM) for the interaction of information across different layers in the Transformer decoder, guiding the model to better distinguish equipment features. Secondly, we adopt the cascade architecture to achieve multi-stage refinement of features. Experimental results demonstrate that our proposed CQE-Mask DINO achieves detection accuracy of 60.5% and segmentation accuracy of 64.5% on the test dataset. Compared to the baseline model, the accuracy is improved by 5.8% and 2.1% respectively. Furthermore, CQE-Mask DINO outperforms existing state-of-the-art models in both object detection and instance segmentation tasks.

Visual anomaly detection includes image anomaly detection and video anomaly detection, focusing on identifying and locating anomalous patterns or events in images or videos. This technology finds widespread applications across multiple domains, including industrial surface defect inspection, medical image lesion analysis, and security surveillance systems. By identifying patterns that do not conform to normal conditions, it helps to detect anomalies in a timely manner and reduce risks and losses. This paper provides a comprehensive review of existing visual anomaly detection algorithms. It introduces a taxonomy of algorithms from a new perspective: statistical-based algorithms, measurement-based algorithms, generative-based algorithms, and representation-based algorithms. Furthermore, this paper systematically introduces datasets for visual anomaly detection and compares the performance of various algorithms on different datasets under typical evaluation metrics. By analyzing existing algorithms, we identify current challenges and suggest promising future research directions.

The acquisition of phenotype parameters with computer vision is crucial for smart breeding, cultivation management, and automated harvesting. However, occlusion in Oudemansiella raphanipies hinders accurate segmentation and phenotype information collection. This study proposes ORP-extractor (Oudemansiella raphanipies phenotype extractor), a deep learning model designed to address the above-mentioned challenges. Initially, to realize instance segmentation of individual Oudemansiella raphanipies and acquired its complete shape, a newly improved Mask R-CNN networks (named OR R-CNN) was designed, which integrated the advantages of the Cross-Criss attention module and PointNet. Furthermore, with the shape prior of the cap-stem contour, an automatic measurement-position search method was proposed to assist in phenotype parameter extraction. Finally, four phenotypic parameters (cap diameter, cap height, stem diameter and stem length) were calculated combining the measurement positions with depth image. In addition, to increase the accuracy of annotation and save cost, a novel occlusion image synthesis strategy for ORP-extractor training also introduced. The segmentation results showed an AP of 86.58

Radar emitter individual identification is one of the key technologies of modern electronic countermeasure reconnaissance and electronic intelligence. With the advancement of radar technology and the increasingly complex electromagnetic environment, existing methods for identifying emitter are gradually becoming unable to meet the performance requirements of modern radar individual identification. Aiming at improving the adaptability of feature extraction for non-cooperative radar emitter signals and the robustness of individual identification in the complex modern electronic warfare environment, a signal fingerprint feature extraction method based on decomposition and fusion is proposed. It firstly integrates signal decomposition and scattering convolution networks (SCN) to adaptively extract the multi-scale intra-pulse feature of the signal, while removing the potential noise of the redundant component by energy proportion. And then a deep feature fusion model based on multi-head self-attention and residual connection is proposed to fuse the multi-scale features and the time domain features to further extract signal fingerprint of radar emitter. Experimental results based on the real radar emitter signals demonstrate that the identification method proposed in this paper can more effectively extract signal fingerprint features and the identification accuracy reaches 96.45%, which outperforms other existing identification methods.

Action recognition is a promising task of identifying human activities in videos or images. Human movement is often accompanied by occlusion and blurring, resulting in many approximate behaviors that cannot be correctly recognized. Most methods directly employ local or motion cues to improve global features. Due to insufficient exploration of 3-D depth information, similar actions from the 2-D perspective still cannot be particularity distinguished. To tackle this challenge, this article proposes a novel action recognition approach termed RPS-Net, integrating 3-D human body reconstruction and hybrid pose self-supervision (HPSS). The designed 3-D reconstruction network is named CFFormer, which leveraging a context-fusion structure to generate 3-D meshes as auxiliary input. Among them, the mesh rotated 90 degrees (M9) is exploited to provide multiperspective associative information, which employing spatial residual learning to enhance extra 3-D cues under various interpose variations. Meanwhile, the original images and meshes from the same perspective will perform HPSS with temporal encoding. It is responsible for capturing subtle differences in key point positions and structures across multiple dimensions. Extensive experiments demonstrate that our proposed method significantly surpasses most existing algorithms. And the Top-1 recognition accuracy on SSV2, HMDB51, and Olympic Sports datasets can reach 73.4%, 87.7%, and 96.2%, respectively.

This paper proposes a Siamese motion-aware Spatio-temporal network ( SiamMAST ) for video action recognition. The SiamMAST is designed based on the fusion of four features via processing video frames: spatial features, temporal features, spatial dynamic features, and temporal dynamic features of a moving target. The SiamMAST comprises AlexNets as the backbone, LSTMs, and the spatial motion-awareness and temporal motion-awareness sub-modules. RGB images are fed into the network, where AlexNets extract spatial features. Further, they are fed into LSTMs to generate temporal features. Additionally, spatial motion-awareness and temporal motion-awareness sub-modules are proposed to capture spatial and temporal dynamic features. Finally, all features are fused and fed into the classification layer. The final recognition result is produced by averaging the test label probabilities across a fixed number of RGB frames and selecting the label of the highest probability. The whole network is trained offline using an end-to-end approach with large-scale image datasets using the standard SGD algorithm with back-propagation. The proposed network is evaluated on two challenging datasets UCF101 (93.53%) and HMDB51 (69.36%). The experiments have demonstrated the effectiveness and efficiency of our proposed SiamMAST .

Three-dimensional human pose estimation focuses on generating 3D pose sequences from 2D videos. It has enormous potential in the fields of human–robot interaction, remote sensing, virtual reality, and computer vision. Existing excellent methods primarily focus on exploring spatial or temporal encoding to achieve 3D pose inference. However, various architectures exploit the independent effects of spatial and temporal cues on 3D pose estimation, while neglecting the spatial–temporal synergistic influence. To address this issue, this paper proposes a novel 3D pose estimation method with a dual-adaptive spatial–temporal former (DASTFormer) and additional supervised training. The DASTFormer contains attention-adaptive (AtA) and pure-adaptive (PuA) modes, which will enhance pose inference from 2D to 3D by adaptively learning spatial–temporal effects, considering both their cooperative and independent influences. In addition, an additional supervised training with batch variance loss is proposed in this work. Different from common training strategy, a two-round parameter update is conducted on the same batch data. Not only can it better explore the potential relationship between spatial–temporal encoding and 3D poses, but it can also alleviate the batch size limitations imposed by graphics cards on transformer-based frameworks. Extensive experimental results show that the proposed method significantly outperforms most state-of-the-art approaches on Human3.6 and HumanEVA datasets.

The popularity of Agrocybe cylindracea is increasing due to its unique flavor and nutritional value. The Agrocybe cylindracea cap is a key aspect of the growth process, and high-throughput observation of cap traits in greenhouses by machine vision is a future development trend of smart agriculture. Nevertheless, the segmentation of the Agrocybe cylindracea cap is extremely challenging due to its similarity in color to the rest of the mushroom and the occurrence of mutual occlusion, presenting a major obstacle for the effective application of automation technology. To address this issue, we propose an improved instance segmentation network called Agrocybe cylindracea R-CNN (AC R-CNN) based on the Mask R-CNN model. AC R-CNN incorporates hybrid dilated convolution (HDC) and attention modules into the feature extraction backbone network to enhance the segmentation of adhesive mushroom caps and focus on the segmentation objects. Furthermore, the Mask Branch module is replaced with PointRend to improve the network’s segmentation accuracy at the edges of the mushroom caps. These modifications effectively solve the problems of the original algorithm’s inability to segment adhesive Agrocybe cylindracea caps and low accuracy in edge segmentation. The experimental results demonstrate that AC R-CNN outperforms the original Mask R-CNN in terms of segmentation performance. The average precision (AP) is improved by 12.1%, and the F1 score is improved by 13.7%. Additionally, AC R-CNN outperforms other networks such as Mask Scoring R-CNN and BlendMask. Therefore, the research findings of this study can meet the high-precision segmentation requirements of Agrocybe cylindracea caps and lay a theoretical foundation for the development of subsequent intelligent phenotyping devices and harvesting robots.

The arcing happened in Pantograph-catenary system could damage railway operation. Arcing are characterized by short and volatile durations, making it challenging to rapidly detect and segment arcing of varying shapes and sizes. To address these issues, a fast deep learning based semantic segmentation model named ECA-ArcSC is proposed for real-time arcing detecting and segmentation task. Firstly, four downsampling convolution layers are utilized to capture spatial information of the arcing. Concurrently, a four-stage downsampling strategy is employed to calculate contextual information of the arcing. Finally, the spatial and contextual information of the arcing is aggregated to predict the final dense semantic segmentation output. Experimental results on the Pantograph-catenary arcing dataset and comparative analysis demonstrate the effectiveness of the proposed approach, achieving IoU of 76.08% and Recall of 84.3%. For inputs of size 512×512, ECA-ArcSC achieves FPS of 123 on an NVIDIA 4070Ti graphics card.

Mushroom cap is a key trait in the growth process and its phenotypic parameters are essential for automatic cultivation and smart breeding. However, the edible mushrooms are usually grown densely with mutual occlusion, which is difficult to obtain the phenotypic parameters non-destructively. Although deep learning methods achieve impressive performance with superior generalization capabilities, they require a large amount of ground truth label of the occluded target, which is a challenging task due to the substantial workload. To overcome this obstacle, a novel synthetic cap occlusion image method was proposed for rapidly generating edible mushroom occlusion datasets using raw images, in which the ground truth is obtained from the real world and the occlusion is randomly generated for simulating real scenes. Moreover, variants of amodal instance segmentation models with different backbone were trained and evaluated on our synthetic occlusion image datasets. Finally, an amodal mask-based size estimation method was presented to calculate the width and length of the cap. The experimental results showed that the amodal instance segmentation achieved an AP@[0.5:0.95] of 82%, 93% and 96% on Oudemansiella raphanipes, Agrocybe cylindraceas and Pholiota nameko synthetic cap datasets, respectively, with a size of 1024 × 1024 px, which indicates that our occlusion image synthesis method can effectively simulate the real cap occlusion situation. The size estimation method achieved an R2 of 0.95 and 0.98 between predictive amodal caps and manually labeled caps for the length and width of Agrocybe cylindraceas cap, respectively, which can be applied to obtain the phenotypic parameters of each cap effectively and accurately. These methods not only meet the demand for automatic monitoring of edible mushroom morphology in factories but also provide technical support for intelligent breeding.

High-throughput acquisition of phenotypic parameters based on machine vision is important for intelligent breeding, digital cultivation and automated harvesting of Oudemansiella raphanipes. However, due to the occlusion among Oudemansiella raphanipes in the growing bed, it is challenging to accurately and rapidly capture their full shape with conventional methods, resulting in a low measurement success rate. The overall goal of this study is to propose a deep learning-based method (Oudemansiella raphanipes Occlusion Region-based Convolutional Neural Networks, OR-ORCNN) to obtain the morphology of occluded Oudemansiella raphanipes precisely in the mushroom growing bed with high efficiency. To accomplish this, a state-of-the-art attention architecture (OR-SE), which consists of CCA (Criss-Cross Attention) and MSF (Multiscale Fusion) modules, was applied to the deocclusion assignment. Then, the predicted amodal mask combined pinhole camera model was used to compute the size of Oudemansiella raphanipes caps in millimeters (mm). The experimental results showed an improvement in the AP50 of 2.78 % and mAR of 5.60 % compared with the baseline model, demonstrating the effectiveness of the OR-SE module in improving the feature information capability of backbone networks. Meanwhile, the Oudemansiella raphanipes cap size estimation results reported an MAE of 0.93 mm and a MAPE of 4.96 %. Furthermore, to confirm the robustness and generalizability of the OR-ORCNN model, we also conducted experiments on Agrocybe cylindraceas and obtained satisfactory results. In summary, the proposed algorithm provides an effective way to recover the shape of the occluded region of Oudemansiella raphanipes with high efficiency and precision, which can be used to help researchers breed intelligently and growers optimize management.

To address the shortcomings of the widely-linear complex-valued distributed adaptive filtering algorithm that cannot combine convergence speed and steady-state performance when using a fixed step-size strategy. In this paper, by making the partial derivative of the square of the sum of a priori error and a posteriori error with respect to step-size equal to zero, we propose a novel variable step-size (VSS) strategy, which is formulated as a function of a priori error and a posteriori error. In addition, a corresponding adaptive filtering algorithm is proposed by using the VSS strategy. Then in the process of parameter update, the moving average method is used to avoid large perturbations in iteration of the algorithm. Secondly, the theoretically analysis of the transient and steady-state performance of the proposed algorithm is provided. Finally, in the numerical simulation experiments, the effect of each parameter on the algorithm is tested, the result of the comparison with the fixed-step algorithm illustrates that the proposed algorithm has more superiority, and the high degree of matching between theory and experiment also verifies the feasibility and accuracy of the theoretical analysis method.