Semantic Segmentation of Corrosion in Cargo Containers Using Deep Learning

As global trade expands, container terminals face growing pressure to improve efficiency and capacity. During the process of loading and unloading containers, several inspections are performed with the urgent need to minimize delays. In this paper we explore corrosion, as it poses a persistent threat that compromises container durability and leads to costly repairs. Identifying this threat is no simple task, as it varies in form, progresses unpredictably, and is influenced by diverse environmental conditions and container types. In collaboration with the Port of Sines, Portugal, this work explores a potential solution for a real-time computer vision system, with the aim to improve container inspections using deep learning algorithms. We propose a system based on the semantic segmentation model, DeepLabv3+, for precise corrosion detection using images provided from the terminal. Given that the data was entirely raw and unprocessed, several techniques were applied for pre-processing, along with a review of various annotation tools. Once the data and annotations were prepared, we explored two approaches: leveraging a pre-trained model originally designed for bridge corrosion detection and fine-tuning a version specifically for cargo container assessment. Achieving performance results of 49% corrosion detection on the fine-tuned model, this work showcases the potential of deep learning in automating inspection processes and highlights the importance of generalization and training in real-world scenarios, exploring innovative solutions for smart gates and terminals.