| Popis: |
We consider an industrial internet of things environment, where involves multiple production factors, such as automatic guided vehicles (AGVs), people, container, etc. A deep learning model is presented for multi-target recognition, where the training data is shadow image formed by the nonuniform illumination of LED lighting source. Three shadow models of typical shapes are constructed to describe the shadows at different positions. The performance of the optimal VGG-16-based Faster-RCNN model is analyzed in view of the recognition accuracy and speed, and it is proved that recognizing three, four, and five types of objects, the mean average precision is 93%, 94.8%, and 92.6%, respectively. To enhance the generalization performance, the optimal Faster-RCNN is combined with the motion state of objects and the corresponding threshold. Simulation results show that the proposed deep learning model obtains significant performance gains to reduce missed and false detection. |
