To solve the problem of insufficient feature fusion between the deep and shallow feature layers of the original YOLOX algorithm, which resulting in a loss of object semantic information, this paper proposes a YOLOX object detection algorithm based on attention and bidirectional cross-scale path aggregation. First, an efficient channel attention module is embedded in the YOLOX backbone network to reinforce the key features in the object region by distinguishing between the importance of the different channels in the feature layer, thus enhancing the detection accuracy of the network. Second, a bidirectional cross-scale path aggregation network is designed to change the information fusion circulation path while increasing the cross-scale connections. Weighted feature fusion is used to learn the importance of the different path input features for differentiated fusion, thereby improving the feature information fusion capability between the deep and shallow layers. Finally, the SIOU loss function is introduced to improve the detection performance of the network. The experimental results show that on the PASCAL VOC2007 and MS COCO2017 datasets, the algorithm in this paper improves mAP by 2.32% and 1.53% compared with the original YOLOX algorithm, and has comprehensive performance advantages compared with other algorithms. The mAP reaches 99.44% on the self-built iron ore metal foreign matter dataset, with a recognition speed of 56.90 frames/s.

Action assessment and feedback can effectively assist fitness practitioners in improving exercise benefits. In this paper, we address key challenges in human action recognition and assessment by proposing innovative methods that enhance performance while reducing computational complexity. Firstly, we present Oct-MobileNet, a lightweight backbone network, to overcome the limitations of the traditional OpenPose algorithm’s VGG19 network, which exhibits a large parameter size and high device requirements. Oct-MobileNet employs octave convolution and attention mechanisms to improve the extraction of high-frequency features from the human body contour, resulting in enhanced accuracy with reduced model computational burden. Furthermore, we introduce a novel approach for action recognition that combines skeleton-based information and multiple feature fusion. By extracting spatial geometric and temporal characteristics from actions, we employ a sliding window algorithm to integrate these features. Experimental results show the effectiveness of our approach, demonstrating its ability to accurately recognize and classify various human actions. Additionally, we address the evaluation of traditional fitness exercises, specifically focusing on the BaDunJin movements. We propose a multimodal information-based assessment method that combines pose detection and keypoint analysis. Label sequences are obtained through a pose detector and each frame’s keypoint coordinates are represented as pose vectors. Leveraging multimodal information, including label sequences and pose vectors, we explore action similarity and perform quantitative evaluations to help exercisers assess the quality of their exercise performance.