IDENTIFICATION AND CLASSIFICATION OF BANANA RIPENESS STAGE BASED ON COLOR SPACE PARAMETERS USING SUPPORT VECTOR MACHINE, DECISION TREE, DAN RANDOM FOREST CLASSIFIER

• Main Author: Hasani Siregar, Yusnan
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/43875
• Institution: Institut Teknologi Bandung

Banana ripening can be observed from several features such as the changing of the color, texture, and chemical properties. During the ripening, banana color is changing from mostly green to mostly yellow due to the decreasing number of the chlorophyll. To predict banana ripeness level, the existing methods use visual observation by comparing the banana color with a standard chart color. This method is unreliable since it depends on the observer’s judgment and the lighting condition. In this study, classification algorithms was developed to predict banana ripeness level by analyzing the changing pattern of the banana color space. First, dataset created from 12 pieces of bananas by taking the picture of the ripening bananas. Next, we use threshold segmentation to crop the banana from the background image. Thereafter, the dataset can be extracted from each color space. We generate the dataset from four color-space: RGB, HSV, L*a*b*, and LCH. We consider two classification modes: all color-space and per color-space classification. As a comparison, we implement the ripeness prediction using four classifier models: Linear support-vector-machine (SVM-L), Radial support-vector-machine (SVM-R), decision tree (DT), and random forest (RF). We use 70 % of dataset as the training data and the rest (30%) as the testing data. To achieve high confidential interval, we rerun each classifier for 1000 times. For the performance evaluation, we use confusion matrix analysis with 33 images. The result shows that for all color-space classification SVM-L classifiers perform the best with the accuracy of 85,64 ± 5,26 %. For per color-space classification, the best accuracies are performed by SVM-L classifier in Lch and Lab color space with the accuracy of 86,74% ± 5,22% and 85,94% ± 5,49%. SVM-R classifier performs the worst for both classification modes, all color-space and per color-space classification.