Image recognition method for stereoscopic warehouse checking of tobacco enterprise

The invention discloses an image recognition method for tobacco enterprise stereoscopic warehouse checking. The method is a hybrid network model based on double-layer routing attention improvement. In the inventory checking process, an image data set of finished cigarette inventory in a stereoscopic warehouse of a tobacco enterprise is obtained through shooting equipment, the collected image data set of the finished cigarette inventory is expanded through preprocessing, and multi-scale feature extraction is conducted on finished cigarette pictures through an improved hybrid network; and finally, an image classification task of finished cigarettes in a warehouse is completed through product multi-scale features extracted by the network. The network model used in the invention can extract multi-scale features and is relatively simple in structure, and the intelligent inventory checking task of the stereoscopic warehouse inventory of the tobacco enterprise is realized with higher precision and efficiency; throug