Recording the trajectory of table tennis balls in real-time enables the analysis of the opponent’s attacking characteristics and weaknesses. The current analysis of the ball paths mainly relied on human viewing, which lacked certain theoretical data support. In order to solve the problem of the lack of objective data analysis in the research of table tennis competition, a target detection algorithm-based table tennis trajectory extraction network was proposed to record the trajectory of the table tennis movement in video. The network improved the feature reuse rate in order to achieve a lightweight network and enhance the detection accuracy. The core of the network was the “feature store & return” module, which could store the output of the current network layer and pass the features to the input of the network layer at the next moment to achieve efficient reuse of the features. In this module, the Transformer model was used to secondarily process the features, build the global association information, and enhance the feature richness of the feature map. According to the designed experiments, the detection accuracy of the network was 96.8% for table tennis and 89.1% for target localization. Moreover, the parameter size of the model was only 7.68 MB, and the detection frame rate could reach 634.19 FPS using the hardware for the tests. In summary, the network designed in this paper has the characteristics of both lightweight and high precision in table tennis detection, and the performance of the proposed model significantly outperforms that of the existing models.

To address the challenges posed by complex backgrounds and the low occurrence in photovoltaic cell images captured by industrial sensors, we propose a novel defect detection method: MRA-YOLOv8. First, a multi-branch coordinate attention network (MBCANet) is introduced into the backbone. The coordinate attention network (CANet) is incorporated to mitigate the noise impact of background information on the detection task, and multiple branches are employed to enhance the model’s feature extraction capability. Second, we integrate a multi-path feature extraction module, ResBlock, into the neck. This module provides finer-grained multi-scale features, improving feature extraction from complex backgrounds and enhancing the model’s robustness. Finally, we implement alpha-minimum point distance-based IoU (AMPDIoU) to the head. This loss function enhances the accuracy and robustness of small object detection by integrating minimum point distance-based IoU (MPDIoU) and Alpha-IoU methods. The results demonstrate that MRA-YOLOv8 outperforms other mainstream methods in detection performance. On the photovoltaic electroluminescence anomaly detection (PVEL-AD) dataset, the proposed method achieves a mAP₅₀ of 91.7%, representing an improvement of 3.1% over YOLOv8 and 16.1% over detection transformer (DETR). On the SPDI dataset, our method achieves a mAP₅₀ of 69.3%, showing a 2.1% improvement over YOLOv8 and a 6.6% improvement over DETR. The proposed MRA-YOLOv8 also exhibits great deployment potential. It can be effectively integrated with drone-based inspection systems, allowing for efficient and accurate PV plant inspections. Moreover, to tackle the issue of data imbalance, we propose generating synthetic defect data via generative adversarial networks (GANs), which can supplement the limited defect samples and improve the model’s generalization ability.

Advancements in deep learning and computer vision have led to the discovery of numerous effective solutions to challenging problems in the field of agricultural automation. With the aim to improve the detection precision in the autonomous harvesting process of green asparagus, in this article, we proposed the DA-Mask RCNN model, which utilizes the depth information in the region proposal network. Firstly, the deep residual network and feature pyramid network were combined to form the backbone network. Secondly, the DA-Mask RCNN model added a depth filter to aid the softmax function in anchor classification. Afterwards, the region proposals were further processed by the detection head unit. The training and test images were mainly acquired from different regions in the basin of the Yangtze River. During the capturing process, various weather and illumination conditions were taken into account, including sunny weather, sunny but overshadowed conditions, cloudy weather, and daytime greenhouse conditions as well as nighttime greenhouse conditions. Performance experiments, comparison experiments, and ablation experiments were carried out using the five constructed datasets to verify the effectiveness of the proposed model. Precision, recall, and F1-score values were applied to evaluate the performances of different approaches. The overall experimental results demonstrate that the balance of the precision and speed of the proposed DA-Mask RCNN model outperform those of existing algorithms.

Recognizing behaviors within classroom settings is vital for gauging educational progress and optimizing teaching methodologies. The complexity of classroom environments often poses challenges for detectors in terms of detection performance. Despite having a simplified structure for analyzing actions in classroom environments, detection transformer (DETR) encounters certain challenges. To overcome these obstacles, we propose a novel real-time end-to-end object detection model called PACR-DETR. Specifically, we introduce an efficient position-aware channel residual network (PACRNet), which excels at extracting information from deep layers along the spatial dimension within complex backgrounds. Notably, it exclusively employs the transformer encoder to process the deepest layer of the feature map. Afterwards, we introduce the efficient residual mixing block (ERM Block), in order to apply convolution operations selectively to a subset of channels. This block underpins the multi-branch perception cross-level fusion module (MPCF Module), which facilitates hierarchical and cross-layer information fusion. Through a bidirectional chain, the multi-branch perception cross-level fusion network (MPCFNet) is architected to comprehensively integrate multi-scale gradient information. Finally, we integrate the normalized Wasserstein distance (NWD) with the Inner-MPDIoU loss function to enhance the recognition performance for small-scale objects. The results show that the PACR-DETR outperforms other methods in detection performance and real-time capability. On the SCB03-S dataset for classroom behavior detection against complex backgrounds, the proposed method attains a precision of 72.1%, outperforming YOLOv6 by 14.1% and YOLOv8m by 4.5%. Furthermore, it reaches 41.67 FPS, meeting the requirements for real-time application.

Under a student-centered educational paradigm, project-based learning (PBL) assessment requires accurate identification of classroom behaviors to facilitate effective teaching evaluations and the implementation of personalized learning strategies. The increasing use of visual and multi-modal sensors in smart classrooms has made it possible to continuously capture rich behavioral data. However, challenges such as lighting variations, occlusions, and diverse behaviors complicate sensor-based behavior analysis. To address these issues, we introduce CSSA-YOLO, a novel detection network that incorporates cross-scale feature optimization. First, we establish a C2fs module that captures spatiotemporal dependencies in small-scale actions such as hand-raising through hierarchical window attention. Second, a Shuffle Attention mechanism is then integrated into the neck to suppress interference from complex backgrounds, thereby enhancing the model’s ability to focus on relevant features. Finally, to further enhance the network’s ability to detect small targets and complex boundary behaviors, we utilize the WIoU loss function, which dynamically weights gradients to optimize the localization accuracy of occluded targets. Experiments involving the SCB03-S dataset showed that CSSA-YOLO outperforms traditional methods, achieving an mAP₅₀ of 76.0%, surpassing YOLOv8m by 1.2%, particularly in complex background and occlusion scenarios. Furthermore, it reaches 78.31 FPS, meeting the requirements for real-time application. This study offers a reliable solution for precise behavior recognition in classroom settings, supporting the development of intelligent education systems.