Design And Implementation Of Purple Soil Classification System Based On Siamese Network

Purple soil is an important tillage soil in southwest China.For ordinary agricultural practitioners,identifying soil types is the basis for land cultivation and crop cultivation.The purpose of this paper is to design and implement a purple soil classification and recognition system to obtain the auxiliary system of purple soil classification and recognition.The main work of this paper is as follows:(1)Image segmentation based on texture-based superpixel algorithm.For traditional super pixel algorithm caused by only considering the characteristics of the distance and color segmentation accuracy is not high,the introduction of the LBP texture characteristics,through the adaptive selection of appropriate threshold to extract the contour information combined with the feature of distance and color,weighted,the images,get through post-processing algorithm(outlier eliminating algorithm and hole filling algorithm),after get the final segmentation image.(2)Model construction based on twin network.Classical and popular models cannot be directly applied to purple soil images,and the traditional convolutional neural network structure is a single chain structure with low precision.In order to solve these problems,the convolutional neural network is designed as twin structure.Based on the input of RGB and Lab color space,the integrated purple soil features were obtained,and the classification results were obtained through the output of the full connection layer.(3)Purple soil classification and identification system for general agricultural workers.This system in the Window based on C++ language development environment,MFC application framework,algorithm execution language for Matlab,Microsoft SQL Server data platform.The interface and functions of the system are designed and implemented through VS program.The interface is considered practical and simple.The main requirements of the system are to realize the interface of sample training curve,the functional module of user operation log and user management,the reality interface after pre-segmentation,and the display interface of classification results.