Optimized Deep Neural Network Architectures with Anchor Box optimization for Shipping Container Corrosion Inspection

Abstract

Shipping containers have revolutionized the way cargo is transported all around the world by ensuring the safety of shipments during their operating lives. When containers are in transit all year, they age and are exposed to different weather conditions. Therefore, containers are susceptible to deterioration and damages. The corroded surface of the shipping container is one of the most essential defects known on the surface of the containers that leads to severe damage such as hole in which more effort and cost are needed to repairing. The existing solution for shipping container corrosion detection is based on visual inspection that requires human experts to manually inspect containers that are time-consuming. To address this problem, in this paper the optimized deep neural network architecture is proposed to automatically inspect corrosion defects on the surface of shipping containers. In the proposed architecture, deep neural network models including Faster R-CNN, SSD-Mobile net, and SSD Inception V2 are employed and optimized with anchor box optimization to inspect the corrosion defect and localize it on the surface of shipping containers. The accuracy and speed of the deep neural networks in inspecting corrosion defects on shipping containers are compared and analyzed in experimental results. The experimental results demonstrate that the combination of deep neural networks and anchor box optimization improves the performance of models in detecting corrosion of the surface of shipping containers.