Combining Discriminant Analysis and Neural Networks for Detection of Internal Defects in Mangoes using X-Ray Imaging Technique

Abstract

—In today’s competitive world, quality is considered as the key factor in the modern food industry and the quality of agricultural produce is of main concern for export. Specifically, quality of fruits is of major concern in the export and import industry as it has to conform to the quality norms of the corresponding country. In recent years, non-invasive imaging techniques such as Magnetic resonance imaging (MRI), X-ray, Computed tomography (CT), Nuclear magnetic resonance (NMR), Near infrared (NIR), Ultrasound and Hyper-spectral imaging are being employed to determine the quality of fruits. The “king of fruits”, Mango (Magnifera indica Linn) is the most economically important agricultural crop. India being the major producer of mangoes (50% of global production)and contributing majority of mango cultivars to the world market needs economical, non-destructive methods for quality evaluation of mangoes. There is a need to develop a nondestructive system that objectively classifies the internal quality of mangoes in real time. In this paper, an X-ray based computer vision methodology is proposed to automatically detect internal defects of mangoes and classify the quality into two groups, “Defective” and “Non-defective”. In the proposed methodology we built a dataset of 572 X-ray images of mangoes and validated it using Discriminant Function Analysis (DFA) predictive model which determines the group membership of each sample in the dataset based on the huge feature space extracted from the sample images. The features that best predicts the group membership were given as inputs to Multilayer PerceptronNeural Network (MLP NN) with scaled conjugate gradient optimization algorithm and the optimized MLP architecture with maximum classification accuracy was determined. The proposed model was able to classify the X-ray image samples into Defective and Non-defective groups with an accuracy of 91.3%.