Research On Mineral Recognition And Mobile Application Based On Machine Vision

With the development of the national requirements for green mining and smart mining,mineral identification,as an important basic work for mining and geological research,has become a major challenge for efficient and accurate identification of minerals and practical applications.Traditional mineral identification methods suffer from the problems of complexity,cumbersome operation procedures,and low efficiency.In this paper,we propose an improved ConvNeXt network-based mineral identification model,which effectively improves the performance and efficiency in identifying different minerals based on their morphological characteristics.Furthermore,by using knowledge distillation methods and combining software design,the compressed model is applied to mobile devices,which is of great significance for intelligent mineral identification.The main research work is as follows:(1)Construction of Mineral Image Dataset.In this work,we captured images of 26 different minerals using a camera.To address the issue of insufficient sample size,we collected additional samples from the "National Rock and Mineral Specimen Resource Platform" and the mineral database platform "Min Dat".In addition,we used image enhancement algorithms to expand the mineral image dataset,thus constructing a highquality dataset for mineral identification tasks.(2)Construction of Mineral Recognition Model Based on Improved ConvNeXt Network.In this work,we addressed the unique characteristics of minerals such as varying shapes and colors by selecting the ConvNeXt network for deep learning image classification as the basis for mineral recognition.We incorporated the Sim AM module for mineral feature extraction,introduced the powerful channel attention mechanism ECA-Net,added the ASFF feature pyramid module,and utilized transfer learning to optimize and improve the model.We also designed activation functions and optimizers to construct the mineral recognition model.We conducted experiments and training using26 mineral ore images and compared the results from four aspects: training results,model performance,model effectiveness,and mineral image feature visualization.The experimental results showed that the mineral recognition model constructed in this work had the best performance and produced the best results.(3)Mobile Application of Mineral Recognition.In order to solve the problem that mineral recognition cannot be applied to practical scenarios,the issue of large model parameters in the constructed mineral recognition model needs to be addressed first.To this end,this study takes knowledge distillation as the theoretical basis,and establishes a corresponding model compression experimental process to compress the mineral recognition model.The experimental results also show that the compressed model performs better.Secondly,the problem of converting the model to the mobile platform needs to be solved.Based on the Py Torch deep learning network framework,this study converts the intermediate model and designs and tests the mineral recognition software on the Android Studio platform to address the issue of embedding the model in the mobile platform,thereby achieving the application of mineral recognition on mobile devices.This article focuses on the application of deep learning in the identification of minerals using mineral images as the research object.The article builds a mineral identification model,utilizes knowledge distillation,and combines the Pytorch framework and the Android Studio platform to achieve the entire process of applying mineral identification to mobile devices.In summary,this article employs machine vision as a research method,covering the construction and experimental validation of the mineral identification model,as well as the entire process of mineral identification application on mobile devices,thus providing important theoretical value and practical significance in the field of mineral identification.