Classification of Chili Plant Condition based on Color and Texture Features

Plant health conditions can be identified destructively and non-destructively. However, the destructive method was considered ineffective due to human error because of repeated sample tests, limited equipment, queue duration, and reading errors. Non-destructive methods such as digital image processing can be used to determine plant health conditions more quickly and objectively. This study combines two features, color, and texture, based on the statistical characteristics of RGB, Grayscale, and Local Binary Pattern (LBP) images. The results of feature extraction are processed using the Multi-Layer Perceptron learning method. Based on the experiments, the combination of RGB, Grayscale, and LBP features provides the best performance compared to a single feature. In addition, good MLP performance is obtained using three hidden layers with the number of nodes respectively 2048, 512, and 256. MLP can help determine seven plants health conditions with the highest accuracy of 79.67%.