A Comprehensive Deep Learning based Approach for Corrosion Detection of Industrial Machinery and Tools

In the modern era of swift industrial development, the identification of corrosion in machinery and tools is very important to prevent economic loss and ensure operational efficiency. The prevailing manual inspection approaches are labor-intensive and inefficient, often leading to unnoticed flaws that can compromise the safety and functionality of industrial equipment. Thus, the implementation of a smart and automated system is imperative in the industrial sector to enhance quality, accelerate productivity, and ensure consistency. In order to efficiently discover and classify corrosion flaws, we have developed a CORROD dataset that includes corrosion images in several environments. We have also proposed an optimal method that makes use of the most advanced YOLOv5m models. To increase the model’s accuracy, we have also investigated a number of YOLO series, such as YOLOv5s, YOLOv5m, YOLOv5n, YOLOv8s, YOLOv8m, YOLOv8n, and YOLOv9-g-c, and we have performed extensive hyperparameter tweaking. With an average detection time of 4 ms and an impressive Mean Average Precision (mAP) of 96.4% in corrosion detection, the porposed model performs better than all other competing models. Taking into account all these facts, our research has the potential to improve industrial operations and optimize the economy.