Saliency Detection of Logistics Packages Based on Deep Learning

Abstract

Aiming at the problem that parcels in the logistics industry cannot be accurately located on the conveyor belt to obtain parcel location information, this paper proposes a parcel saliency detection and location method based on deep learning. Firstly, RGBD (Red-Green-Blue-Depth) images are acquired by depth cameras, and the images are filtered and hole-filled to remove noise and irrelevant information; then they are input to the constructed neural network model for training and testing; finally, the location of parcels in the images is obtained. Test experiments on the parcels on the conveyor belt show that the accuracy of locating the parcel position reaches 96.92% with an mean absolute error of only 0.0141, which can guarantee the accuracy of locating the parcel position and thus facilitate the post-processing of the parcel information. This method has greater research significance and engineering application value for the logistics industry.