Abstract: In order to find the congestion of the conveyor, the pharmaceutical company needs to locate the medicine boxes and vacancies on the conveyor, but the manual method is of inefficiency and has poor real-time performance. In this context, combining with the Faster R-CNN model, a target detection method is proposed. In this method, training set and testing set are constructed based on the conveyor images, then, the training set is processed through the ZFNet convolutional neural network to calculate the convolution characteristics, and the RPN (Region Proposal Network) is used to generate accurate candidate regions. On this basis, the classification and regression based on the Faster R-CNN model are performed on the candidate regions, and the rectangular boxes of the kit and the vacancy are calculated. At last, the trained model is tested by using the testing set to label the target and to calculate the probability. The result shows that the method has good detection accuracy for the conveyor belt target.

Key words: Faster R-CNN, ZFNet convolutional neural network, target detection