Automatic Defect Detection for Mango Fruit Using Non-Extensive Entropy with Gaussian Gain

Kotchakorn Tiemtud, Pornpimon Saprasert, Thanikarn Tormo, Saifon Chaturantabut

Abstract

Quality inspection process of agricultural products recently becomes a crucial part of food industries. As the amount of these products gets larger, manual quality control based on traditional visual inspection performed by human can be tedious, time-consuming, labor-intensive, and inconsistent. Machine leaning can be used to automate the quality inspection, which will make this process more rapid, consistent, accurate and cost-e cient. This work focuses on identifying defects on mango surfaces. This work applies non-extensive entropy with Gaussian modeling, which is an unsupervised automated texture defect detection technique and therefore it does not require any training image samples in advance. The numerical results from this approach are shown to be e ective in detecting various defects, such as cracks, dark spots and bruises from mango image samples. This work also investigates di erent window sizes used in entropy computation, which will a ect the trade-o between computational speed and detection accuracy.