Study on scrap steel classification using deep learning

The classification of scrap steel is the key step in the recycling and utilization of scrap steel. Human detection has been widely used in the classification of scrap steel carriages at present. One major drawback of this approach is the low efficiency of recycling due to the instability of the operator. Therefore, it is necessary to develop a fast and accurate method for automatic classification of scrap steel carriages. This paper proposes an improving method of classifying scrap steel carriages based on deep learning. First, the key frames in the video stream are obtained by the target detection algorithm, then the features of interests are extracted by the feature extraction algorithm, and finally the classification result of the entire carriage is output by the feature fusion algorithm. In the YOLO algorithm for detecting targets, the Darknet network is abandoned and the MobileNet network is used. The spatiotemporal information separation strategy is used when extracting features. The n×1×1 convolution kernel operator is used in the 3D convolutional network of fusion features. In the self-attention network, only the attention mechanism is set for the time dimension. With the analysis of the different sample ratios of the training set and test set, the method proposed in this paper has the characteristics of strong generalization ability, high accuracy, and fast speed which has provided a deeper insight into classification of scrap steel carriages.