Research on small target garbage detection method based on improved YOLOv7

A DFST-YOLO (Detection for small targets-YOLO) garbage detection model with improved YOLO v7 model is proposed to address the problems of small target size, susceptibility to background interference, weak feature information, and easy to produce missed and false detection in garbage classification detection tasks to be detected. By introducing Omni-Dimensional Dynamic Convolution and Parameter-Free Attention Module in the backbone network, the computing power of the convolutional neural network is improved and the ability to capture key information is enhanced. On the self-built garbage dataset Trash_dataset, mAP
0.5
and mAP
0.5:0.95
reach 95.15% and 70.71%, respectively. The experimental results show that the detection for small targets possesses better detection results compared to some popular algorithms.