Clasificación y detección de defectos en la superficie de las papas utilizando procesamiento digital de imágenes y machine learning.

Abstract

The objective of the curricular integration work was to classify and detect defects on the surface of potatoes using digital image processing and Machine Learning. The classification was based on the standard of the Ecuadorian Standardization Service (INEN) 1516, which indicates that the classification is according to the size of the major axis of the potato considering four categories, and by the absence of damage or defects. This was achieved through digital image processing consisting of image capture, preprocessing, segmentation, feature extraction, and image interpretation steps. For compliance, the first stage used an Emeet C970L webcam placed on a dark box made of aluminum tol where LED lights were also installed to avoid shadows around the potato and improve the quality of the images. The following stages were developed using the Matlab software with the help of different functions, once the potatoes were classified in the different categories, the Machine Learning Softmax regression model was used to improve the efficiency of the system. The results were obtained by comparing the values of the potatoes, obtained by the system and the observation, in addition to the manual measurement of the potato with a caliper king of foot, of which 90% accuracy was obtained with the confusion matrix for the prediction of good and bad potatoes. It is concluded that the system can correctly predict the size of the potato and the defects it has, however, it is recommended to find a method to evaluate the two sides of the potato at the same time and that the entry for the capture of images is automatic.