Research On Recognition Of Compost Maturity Using Machine Vision

Compost maturity is mainly judged by complicated chemical and biological methods,which are cumbersome and inefficient.This paper aims to study the changes of compost appearance using machine vision,so as to realize rapid and accurate judgment of compost maturity.In order to avoid defects in insufficiencies of the number,type,and representativeness of compost images,we collected a large number of compost images according to the origin of compost and the different materials.Finally,nearly 30,000 samples of compost images were used in the experiment.The materials of compost samples included vegetable residue,straw and livestock manure.The samples presented diversity and richness.In this paper,visual-based compost maturity recognition was studied through two technical routes:improved feature fusion with classical feature extraction algorithm and deep learning technique.The main contents are as follows:(1)Texture and color features of compost image are important information for judging whether compost is mature or not.The texture features of the compost images were extracted by Local Binary Pattern(LBP),and the color moment described the color features of the compost.The two features were fused,and a compost maturity recognition model based on LBP-color moments was proposed.Due to texture and color are two completely different features,direct feature fusion led to the suppression of the two image features.It was proposed to fuse the two characteristics of color and texture of compost image by weighting coefficient,finding the best weighting coefficient combination through traversal searching.Features were reduced in dimension using Principal Component Analysis(PCA).The experimental results show that the improved method is superior to the existing composting image based on composting image and other classical and efficient image feature extraction algorithms.(2)Deep learning technology is one of the mainstream artificial intelligence technologies.Among them,Convolutional Neural Network(CNN)is an outstanding representative algorithm for solving image intelligent recognition and classification.CNN has achieved excellent results in agricultural problems such as plant pest detection,fruit maturity classification,and grain damage identification.This paper proposed to apply CNN to compost maturity recognition research,using 3CP-2F,AlexNet,VGG16 and ResNet18 models to carry out experiments on straw,vegetable residues,livestock manure and three compost images mixed data sets,and all of these models had achieved excellent results.Especially,the average accuracies of 3CP-2F and ResNet18 on the four sets of data sets reached more than 99%,and the higher recognition accuracy of compost maturity was obtained.The composting image features of the same raw materials from different sources extracted by CNN models were compared and analyzed.It was proved that the same raw materials from different sources had no impact on the recognition effect of compost maturity.In the application of the model,when the compost raw materials are different,the transfer learning method can be used to further train the network and improve the generalization ability of the compost maturity recognition model.The compost maturity recognition model based on CNN can extract the appearance features of the compost images,and realize the accurate and rapid recognition of compost maturity directly through the compost image under visible light conditions.(3)The compost data automatic collection and compost maturity recognition software were developed.In order to meet the requirements of automatic collection and management of compost images and information of production,the automatic acquisition software for compost production data was designed and developed to realize automatic collection of information such as compost images,production time,compost materials and unit name,without manually collection.It was convenient for users to view and manage compost data.Based on the two models 3CP-2F and ResNet18 with best recognition accuracy,software for compost maturity recognition was developed.Users could select compost images,materials and recognition models to recognize compost maturity just by one button.The compost maturity recognition model compost data acquisition and maturity recognition software meet the requirements of automatic collection of compost images and rapid recognition of compost maturity,and provide reliable guidance and help for compost production.