TOMATO CLASSIFICATION ACCORDING TO ORGANOLEPTIC MATURITY (COLORATION) USING MACHINE LEARNING ALGORITHMS K-NN, MLP, AND K-MEANS CLUSTERING

Abstract

This article presents the design, development, implementation and evaluation of different machine learning type algorithms, for Milano and Chonto tomatoes classification, based on the fruit physical characteristics, such as coloring (maturity degree), taking as reference national and international standards (NTC-1103-1 and USDA, respectively). Different digital image processing techniques are shown, used to describe and extract the characteristics of color statistics of the tomatoes images. For data analysis, supervised and /or trained classification algorithms were implemented with databases and features in the RGB, HSI and L*a*b* color spaces. The techniques for classification used and valued were: K-NN (K-Nearest Neighbors), MLP type Neuronal Networks (Multilayer Perceptron) and unsupervised learning algorithms like K-Means. The evaluation of each classification algorithms is shown, using the global confusion matrix, together with performance indices such as accuracy, precision, sensitivity, and specificity.

Existing System

Conversion to color spaces, luminosity invariability, convolution filters application, regions of interest extraction, histogram analysis, etc.

Proposed System

In the proposed system, the image has been acquired in RGB format, it was necessary to convert it in a gray scale format to facilitate the image pre-processing. Then the images were passed through a filtering stage to clean and reduce of image information, obtaining the characteristic histogram. Once the image was converted to a gray scale format, the quality of the acquired image was improved by applying a non-linear statistical filter. The object pixels of the fruit were separated from the elements that belong to the background, in order to only extract the region of interest to be analyzed. Once the region of interest has been extracted from the scene, the processed image was converted in RGB, HSI, and L*a* b* colors space. Finally, the K-NN algorithm for tomato color classification, together with the MLP neural network method, got the best classification.

Architecture

BLOCK DIAGRAM