A new approach for visual identification of orange varieties using neural networks and metaheuristic algorithms  

作　　者：Sajad Sabzi Yousef Abbaspour-Gilandeh Ginés García-Mateos 

机构地区：[1]Department of Biosystems Engineering,College of Agriculture,University of Mohaghegh Ardabili,Ardabil,Iran [2]Computer Science and Systems Department,University of Murcia,30100 Espinardo,Murcia,Spain

出　　处：《Information Processing in Agriculture》2018年第1期162-172,共11页农业信息处理（英文）

基　　金：This work was partly supported by the Spanish MINECO,as well as European Commission FEDER funds,under grant TIN2015-66972-C5-3-R.

摘　　要：Accurate classification of fruit varieties in processing factories and during post-harvesting applications is a challenge that has been widely studied.This paper presents a novel approach to automatic fruit identification applied to three common varieties of oranges(Citrus sinensis L.),namely Bam,Payvandi and Thomson.A total of 300 color images were used for the experiments,100 samples for each orange variety,which are publicly available.After segmentation,263 parameters,including texture,color and shape features,were extracted from each sample using image processing.Among them,the 6 most effective features were automatically selected by using a hybrid approach consisting of an artificial neural network and particle swarm optimization algorithm(ANN-PSO).Then,three different classifiers were applied and compared:hybrid artificial neural network–artificial bee colony(ANN-ABC);hybrid artificial neural network–harmony search(ANN-HS);and k-nearest neighbors(kNN).The experimental results show that the hybrid approaches outperform the results of kNN.The average correct classification rate of ANN-HS was 94.28%,while ANN-ABS achieved 96.70%accuracy with the available data,contrasting with the 70.9%baseline accuracy of kNN.Thus,this new proposed methodology provides a fast and accurate way to classify multiple fruits varieties,which can be easily implemented in processing factories.The main contribution of this work is that the method can be directly adapted to other use cases,since the selection of the optimal features and the configuration of the neural network are performed automatically using metaheuristic algorithms.

关 键 词：Computer vision Fruits classification Hybrid neural networks Image processing Metaheuristic algorithms 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]