Improve The Research And Application Of YOLOX Algorithm In Small Target Defect Detection

In the field of computer vision,the challenging task of object detection has made breakthrough progress due to the rapid development of deep learning,which is widely used in the application fields of robot vision,autonomous driving,UAV systems and industrial automation.Small target detection has been a difficulty in target detection for a long time.Due to the small size of the target object and the unobvious features after extraction by the network,the traditional target detection model has high leakage detection rate and low detection accuracy in detecting small targets.In practical industrial applications,small target defects and small defects on the product surface,the manual quality detection is extremely inefficient,and it is difficult to control the rate of good products.For traditional image algorithms,they will have poor adaptability and high false alarm rate.So in this paper in the form of deep learning to improve the small target defect detection accuracy,the overall structure in this paper choose Anchor-free model YOLOX model for improvement,using effective layer aggregation network and maximum pooling convolution parallel fusion structure stack,build more efficient backbone feature extraction network,guarantee the extraction features more rich.In the feature fusion stage is more adapted to small target detection multiscale feature fusion way,using four layer scale and context jump connection way,the deep low resolution feature map and high resolution feature map,and combined with Coordinate Attention,to ensure into the detection head features of the high resolution feature layer information more rich.In order to verify the model generality and the effectiveness of small target detection,the public datasets VOC and Visdrone datasets were used for validation,and the experiments show that the improved FS-YOLO has a significant improvement in small target detection accuracy compared with the current mainstream detection model.Utilizing real-time production line data for experimental use,as demonstrated in this paper,holds immense importance in advancing the field of industrial automation.