Surface defect rapid detection method based on lightweight convolutional neural network

Abstract

The invention provides a surface defect rapid detection method YOLOv4-Defect based on a lightweight convolutional neural network, which is used for rapidly and accurately identifying the type of a defect and detecting the position of the defect. The method is based on a YOLOv4 target detection algorithm, on one hand, multi-dimensional compression processing such as convolutional structure optimization, parameter pruning and parameter quantization is performed on a feature extraction network of YOLOv4 to simplify a model and improve the detection efficiency, and the model is pre-trained through knowledge distillation to improve the feature extraction capability of the model, so that the detection precision is improved; and on the other hand, a prediction scale with a more detailed receptive field is added to optimize the model structure, so that the detection performance of the model on tiny defects is improved.

Description

technical field [0001] The invention belongs to the technical field of surface defect detection, and in particular relates to a fast detection method for surface defects of a lightweight convolutional neural network. Background technique [0002] In the industrial production process, it is an essential step to detect defects on the surface of workpieces or products, that is, to identify whether there are defects, the types of defects, and the location of defects. In traditional industries, the detection of workpiece or product surface defects is still in the manual detection stage. Affected by personal factors of workers, it is difficult to guarantee efficiency and quality, and the traditional manual inspection method can no longer meet the needs of the industry. In today's era of explosive growth in data volume and computing power, deep learning has ushered in the spring of development. New deep learning algorithms emerge in endlessly and are widely used in computer vision...

AI Technical Summary

Problems solved by technology

However, the improvement of model performance is accompanied by a substantial increase in the amount of parameters and calculations, which makes the deployment of the model difficult and limited in practical applications, such as portable devices, mobile phones, and high-real-time detection equipment, etc.

[0005] Due to the differences between surface defect detection and other target detection tasks, there are still complex and too large models in the process of applying deep learning technology to surface defect detection, which is not conducive to deployment and application in actual industry, and for Poor detection performance for a variety of tiny surface defects, such as tiny spots and scratches

Images