Abstract:To address the issues of low quality and quantity of samples in ore automatic sorting, a solution combining linear image processing with weighted voting is proposed. The core of this method lies in integrating linear image processing with neural networks to extract image features from multiple aspects. By combining dual-energy subtraction with neural networks, the method enhances the neural network"s ability to distinguish between different density ores, thereby improving the accuracy of sorting. The introduction of the Laplacian operator strengthens the neural network"s detection of image edges, allowing the model to more clearly identify the contours and texture features of the ore. Through multi-network training, the model is able to learn and understand the characteristics of the ore from different perspectives, thus enhancing its generalization ability. In this experiment, a weighted voting mechanism is used to integrate the output results of different networks. During the weighted voting process, each network"s output is assigned a certain weight based on its performance on the validation set. Ultimately, the model determines the classification result for each sample, i.e., concentrate or tailings, based on these weights. Experimental results show that this method has demonstrated extremely high accuracy in the task of ore automatic sorting. The accuracy rates of both dual-energy subtraction technology