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

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.

Machine vision based expert system to estimate orange mass of three varieties

A key issue in fruit export is classification and sorting for acceptable marketing.In the present work,the image processing technique was employed to grade three varieties of oranges(Bam,Khooni and Thompson)separately.The reason for choosing this fruit as the object of the study was its abundant consumption worldwide.In this study,14 parameters were extracted:area,eccentricity,perimeter,length/area,blue value,green value,red value,width,contrast,texture,width/area,width/length,roughness,and length.Further,the ANFIS(Adaptive Network-based Fuzzy Inference System)method was utilized to estimate the orange mass from the data obtained using the image processing in three varieties.In ANFIS model,samples were divided into two sets,one with 70%for training set and the other one with 30%for testing set.The results of the present study demonstrated that the coefficient of determination(R2)of the best model for Bam,Khooni and Thompson measured 0.948,0.99,and 0.98,respectively.In addition,the results indicated that the estimation accuracy of the best model for Bam,Khooni and Thompson was measured as±3.7 g,±1.28 g,±3.2 g,respectively.This result was very satisfactory for the application of ANFIS to estimate the orange mass.