Research on image preprocessing algorithm based on mixed denoising and texture weakening of ore images

Ore particle size information is an important basis for mining enterprises to regulate crushing parameters, due to the complex and harsh environment during the acquisition of ore images on the conveyor belt, resulting in the existence of a variety of composite noise interference in the motion target image, the surface texture characteristics of the ore and the edge of the fuzzy and other problems, thus affecting the effective acquisition of ore particle size information. To address the above issues, an image-denoising network based on global and local feature extraction and an edge enhancement algorithm for texture feature weakening is proposed. The denoising network consists of a shallow local feature extraction module and a Transformer-based U-Net global feature extraction module, which aims to combine the powerful global modeling capability of the Transformer with the local modeling advantage of convolutional network, and reconstructs the image resolution through the dual up-sampling structure, to realize the accurate output of contextual detail information. A texture weakening method based on wavelet transform and fast non-local averaging is proposed to smooth the image and weaken the texture characteristics of the ore surface, and edge sharpening is combined with Bilateral- USMR to enhance the edges of the ore particles to realize the preprocessing of the ore image. The preprocessing results were objectively evaluated and experimentally verified. The results show that the image preprocessing method improves the accuracy of image segmentation and the applicability of the ore particle size measurement technology in complex environments.