Using Artificial Intelligence to Identify Damaged Goods by the External Appearance of the Package when Performing Logistics Operations

Abstract

A proposed promising method for recognising damaged goods by external appearance of packaging using a learning neural network considers the predicted long-term growth in the volume of transported goods, the increasing influence of economic and territorial factors on transportation processes, the complexity of logistics services and the increase in requirements for their quality, the spread of information technology and the improvement of artificial intelligence tools. The objective of the study is to describe the principles of using an artificial neural network to identify damaged goods by their external appearance. The relevance of the problem is confirmed by data on damage to goods during transportation. The methods used in the study help collecting and analysing data, describing and comparing existing and promising technologies, monitoring and modelling the process of cargo handling, and summarising the results. The analysis of common and promising methods of prevention and detection of cargo damage is backed by the results of research on detecting defects on various surfaces and recognising signs and colours in motion using intelligent technologies. The problems of recognising damages on packaging in complex and unfavourable conditions for machine vision are solved with the help of a convolutional neural network. In accordance with the proposed algorithm, image capture is carried out using standard video surveillance cameras. From the image entered into the neural network, fragments with characteristic features are distinguished, which are further checked for compliance with damage patterns. Following damage contour analysis, the neural network recognises the cargo as damaged. The process of training the neural network and integrating the proposed tool throughout the supply chain ensures the recognition of actually damaged goods and elimination of errors associated with minor permissible damage and packaging features. The proposed concept does not require the installation of additional equipment and does not imply a significant cost of damaged cargo recognition services. The paper offers and describes processes of video recording of the cargo flow, loading an image into a neural network, and a model for recognising damaged cargo by the external appearance of the package.