Occlusion target recognition algorithm based on improved YOLOv4

With the development of deep learning, recognition algorithms are increasingly widely used in various fields, and face recognition is a technological embodiment of recognition algorithms in real life. Due to the limited recognition range, the face may be occluded, so it is necessary to design an occluded target recognition algorithm model. This article aims to optimize the “You Only Look Once Version 4” algorithm and propose an improved occlusion target recognition algorithm model by introducing separable convolutional optimization and embedding attention mechanism. This paper designed relevant experiments to verify the model performance and compared the facial recognition model designed by MM Goyani. The experiment shows that the median accuracy of this algorithm and the comparison algorithm are 0.97 and 0.92, respectively, with a distinction of 0.05, and the average values are 0.962 and 0.902, with a discrepancy of 0.060. About the accuracy, the improved algorithm is higher than that unimproved algorithm, with a difference of 16% and an average accuracy difference of 7.5%. Therefore, the constructed algorithm has effectiveness and feasibility, and to a certain extent has good development potential and reference value.