Aiming at the abnormality detection of industrial insert molding processes, a lightweight but effective deep network is developed based on X-ray images in this study. The captured digital radiography (DR) images are firstly fast guide filtered, and then a multi-task detection dataset is constructed using an overlap slice in order to improve the detection of tiny targets. The proposed network is extended from the one-stage target detection method of yolov5 to be applicable to DR defect detection. We adopt the embedded Ghost module to replace the standard convolution to further lighten the model for industrial implementation, and use the transformer module for spatial multi-headed attentional feature extraction to perform improvement on the network for the DR image defect detection. The performance of the proposed method is evaluated by consistent experiments with peer networks, including the classical two-stage method and the newest yolo series. Our method achieves a mAP of 93.6%, which exceeds the second best by 3%, with robustness sufficient to cope with luminance variations and blurred noise, and is more lightweight. We further conducted ablation experiments based on the proposed method to validate the 32% model size reduction owing to the Ghost module and the detection performance enhancing effect of other key modules. Finally, the usability of the proposed method is discussed, including an analysis of the common causes of the missed shots and suggestions for modification. Our proposed method contributes a good reference solution for the inspection of the insert molding process.

Recently, LiDAR-based vehicle-infrastructure cooperative (VIC) perception has shown an advantage in expanding the horizon of connected autonomous vehicles (CAVs), enabling occlusion-aware 3-D scene understanding. Limited communication bandwidth, however, hampers multi-agent cooperation in the urban Internet of Things (IoT) environments. Existing solutions often compress or implicitly filter high-resolution features, resulting in semantic redundancy or information loss, which degrades the overall performance. To tackle this, we propose VICooper, a communication-efficient VIC perception framework. Driven by the stable context of static infrastructure LiDARs, VICooper employs offline background mapping (BgM) to extract foreground points of interest, thereby offering explicit guidance for communication reduction. For the sparse yet critical foreground point clouds, we introduce the multidimensional foreground backbone (MdFB) that incorporates geometric cues, including height, scale, and spatial density, to enrich feature encoding. To address the fusion imbalance between dense vehicle-side and sparse roadside features, we customize a progressive bidirectional feature aggregation (PbFA) using a deformable transformer to capture inter-agent mutual correlations, thereby enabling the deep coupling of heterogeneous agents under asymmetric information. Extensive evaluations on real-world VIC benchmark validate that VICooper achieves superior performance with substantially lower bandwidth, demonstrating its potential in intelligent transportation IoT ecosystems.

Roadside three-dimensional (3-D) object detection is essential for enhancing blind-less perception performance in cooperative autonomous driving. Current research has preliminarily explored the representations from different perspectives and constraints inside individuals. However, the accuracy of existing representations is hardly guaranteed under the wide-range conditions on roadside applications, and the constraints involving multiple targets are less considered. To address these, an accurate representation modeling and interindividual constraint-learning method is proposed. In representation modeling, the deficiencies of existing distance representations are systematically analyzed, and the relative depth is developed with the consideration of numerical distribution and error tendency under wide-range conditions. Besides, the limitation of existing reference-pixel representation is addressed by introducing Affine–Gaussian heatmaps, which accurately selects reference pixels and depresses extra responses based on geometric discrepancies. In constraint learning, the interindividual constraints are effectively extracted in the proposed graph-based module, which introduces powerful graph attention operations and a special designed implicit gradient flow to induce constraints into image feature maps. Extensive experiments on DAIR-V2X-I and Rope3-D demonstrate that significant improvements are achieved compared with concurrent state-of-arts on both the trained scenarios and unseen scenarios.