Faster-YOLO: An Industrial Defect Detection Model Based on Expand Partial Convolution

High-precision detection industrial defect detection models typically incur significant computational costs, rendering it challenging to deploy such models in actual industrial settings where computational resources are often limited. To address this challenge, this paper proposes an industrial defect detection model Faster-YOLO based on expand partial convolution. This model employs the Expand Partial Convolution (EPConv) to reconstruct the C3 module of YOLOv5. Then, the computationally efficient Hard-Swish function is adopted to replace the original SiLU activation function, effectively enhancing the model’s inference speed. Experimental results on different datasets demonstrate that compared to other lightweight models, the proposed model exhibits significant advantages in both inference speed and accuracy, making them more suitable for deployment on low-computational devices.