Empirical Thresholding and Feature Extraction Based Back Propagation- Artificial Neural Network Model for Fruit Grade

Abstract

This paper explains a new feature extraction image pre-processing system followed by back propagation- artificial neural networks based system for class categorization of apple fruit images. Scale Conjugate Gradient (SCG) algorithm is used for back propagation. The methodology comprises of three stages in this work. Firstly, various external image based attributes of apple were taken and process in MATLAB. Size and weight features were also considered as important parameters as only color is not sufficient to judge the quality. Secondly, features extraction was done at image pre-processing for making the algorithm lighter by focusing only key features. Support vector machine (SVM) algorithm is also popular for development of relatively light weight classification models. This work uses artificial neural network toolbox in MATLAB for classification. A single hidden layer BP-ANN (Back propagation- artificial neural network) was used with sigmoid activation functions. The result came in terms of suitable output variable which is the quality class of the apple which is chosen A, B, C and D respectively. The modeling result indicates the tremendous match between the data used in training and assumed output values. It also has shorter calculation time due to the SCG algorithm. It is also possible for apple producers and distributors to classify their fruit using this model and reduce the cost by avoiding manual classification.