Problems of YOLOv3 model

YOLO series object detection network is the most representative network structure in one stage object detection network. YOLOv3 is the latest improved network of YOLO series because the detection accuracy can be comparable to two target detection networks and can achieve real-time detection speed, so it has become one of the most popular object detection algorithms. Considering that the object detection of tomato diseases and pests needs to take into account both the accuracy and speed of detection in practical application, this study takes YOLOv3 as the main body and improves the algorithm according to the application scenario of tomato diseases and pests object detection to complete the location and class identification of tomato diseases and pests.

YOLOv3 network, which has been improved many times, has achieved a good balance between detection accuracy and detection speed and has become the preferred algorithm for many object detection tasks because of its simple implementation. However, as a single object detection network, there are still problems of large positioning error, unbalanced prospects and background complexity.

To pursue detection speed, YOLOv3 algorithm integrates object location and classification into a convolutional neural network, and simultaneously predicts the location coordinates and class information of the object. However, because the deep feature map in the convolutional neural network contains more advanced and abstract feature information, which is suitable for object classification, but because of the loss of more spatial information, it has poor effect on object localization. Shallow feature maps are more specific and contain more spatial information, which is suitable for coordinate positioning but not ideal for object classification. Although YOLOv3 tries to use the concatenation of deep feature map and shallow feature map to fuse different levels of feature information, there is still the problem of inaccurate object location compared with two-stage object detection algorithms.