Research on Nuclear Waste Detection Based on Improved YOLOv3

To address the current problems of nuclear waste detection difficulties and identification rates, this paper proposes to use the improved YOLOv3 for nuclear waste detection. In order to make the algorithm more suitable for the detection of nuclear waste, this paper improves the YOLOv3 algorithm, which replaces the original Darknet-53 backbone network in YOLOv3 with the combination of lightweight network MobileNetV2 and Neighborhood Attention Transformer (NAT). With this method, the size of the weight model will be reduced and the feature scale on the basis of the original algorithm will be decreased, while the detection speed is supposed to be improved. The experimental results show that the improved algorithm will reduce the weight model, decrease average detection time, and increase the mAP _ 0.5. As a result, the misdetection problem of the background sample in YOLOv3 being detected as an article will be overcome. In addition, the rapid and accurate detection of nuclear waste will be realized.