Intelligent monitoring method of tridimensional storage system based on deep learning

Growing international trade requires more flexible warehouse management to match it. In order to achieve more effective warehouse management efficiency, a shelf status-detection method based on deep learning is proposed. Firstly, the image acquisition of a multi-level shelf containing multiple bays is performed under different time and lighting conditions. Due to the difference in image characteristics between the bottom shelf on the ground and the upper shelf on the non-ground level, the collected images were divided into two groups: floor images and shelf images; and the warehouse status recognition was performed on the two groups separately. The two sets of images are cropped and center projection transformed separately to obtain the region of interest. On this basis, the improved residual network model is used to construct different depot detection models for the two sets of images, respectively, and the above algorithm is verified by actual measurements. In this paper, 102,614 images of 3246 depots with different states of non-ground layer, and 27,903 images of ground layer are collected. They are divided into training set and test set according to the ratio of 4:1, and the accuracy of training set is 99.6%, and the accuracy of test set is 99.3%. The experimental outcomes provide a theoretical method and technical support for the intelligent warehouse system management.