| With the rapid development of e-commerce and logistics transportation in recent years,the scale of warehousing is expanding.The supervision problems of storing goods also arise.With the rapid development of artificial intelligence and other technologies,the supervision mode of warehouse goods is also constantly innovating,gradually from artificial management to unmanned,automated management transition.However,there are still problems in some current warehouse goods supervision mode,such as cannot detect and deal with the changes of goods in time and lack the process of monitoring data.Therefore,it is necessary to save monitoring data for a long time.It results in a large storage burden.Aiming at these problems,this thesis takes cotton bale as warehouse goods and designs a warehouse goods supervision system based on deep learning.The main researches are as follows:(1)The method of image rough positioning are proposed and implemented.Considering the complex texture of cotton bale and the influence of noise and illumination difference in monitoring images on feature extraction,this thesis designed a rough positioning processing method based on Sobel and a rough positioning processing method based on HOG to better locate the changed areas in two monitoring images of the same scene.Firstly,the images are preprocessed to remove the influence of noise and illumination difference.Then,the feature operator is used to extract the image features.Finally,the changed regions are roughly located according to the processed feature images.The experimental results show that the accuracy of the rough positioning processing method based on HOG is higher on the test data set established in this thesis,reaching96.0%.And it is taken as the rough positioning processing method used in the system of this thesis.(2)A cotton bale detection and recognition method based on deep learning is proposed and implemented.In order to distinguish whether the changing area is cotton bale accurately and effectively,the feature enhancement network(FEN)module is designed to enrich the feature information by considering the complex texture features of cotton bale,the lack of semantic information and the loss of detail information in network training.And then,two network models for cotton bale detection and recognition are designed,which respectively use the ResNet50 and the VGG16 as the backbone network and combine the FEN module.The experimental results show that the network model using ResNet50 as the backbone network and combining FEN module has higher accuracy on the data set established in this thesis.And it can meet the requirements of practical application.(3)A warehouse goods supervision system based on deep learning is designed and applied to warehouse goods(cotton bale)supervision.This system consists of image acquisition,rough positioning processing,cotton bale detection and recognition based on deep learning and result feedback.The system test proves that the system can find the change of cotton bales in time and feedback the results.In addition,the application of the system can reduce the storage time of monitoring data and effectively solve the problem of the lack of monitoring data processing and storage burden in the traditional supervision method.The system realizes more intelligent,automatic and convenient supervision of cotton bale. |
