Research on Target Detection Technology Based on YOLOv3

Abstract:
Aiming at the low accuracy of YOLOv3 target detection technology in daily detection tasks and the problem that the model is too large, an improved algorithm for changing the network loss function and model pruning is proposed. On the basis of the original YOLOv3 model, the GIoU loss function is used in the location loss part to replace the error square loss to improve the detection accuracy, and the pruning algorithm is used to reduce the redundancy of the model while ensuring that the accuracy is not lost as much as possible. Comparative experiments were carried out on Oxford Hand datasets. Compared with the square error loss, the network detection accuracy rate after using the GIoU loss function is higher. The comparison before and after pruning can effectively reduce the memory and power consumption,and improve calculation efficiency.