Image Optimization Technique Using Local Binary Pattern and Multilayer Perceptron Classification to Identify Potassium Deficiency in Cacao Plants Through Leaf Images

Cocoa plants (Theobroma cacao, L.) are the best plantation crops in Indonesia that play an important role in the economy. However, in cultivation, cocoa farmers often face problems that can cause a decrease in cocoa production, one of which is the lack of potassium nutrients. Therefore, how to implement digital image processing which can help analyze image objects in the form of normal leaf characteristics and potassium-deficient leaf characteristics using the Local Binary Pattern (LBP) method for image feature extraction and classified using the Multilayer Perceptron (MLP) method in identifying potassium deficiencies in cocoa plants based on their leaf characteristics. In the image object in the form of leaf characteristics, each will be identified with 250 in the background dataset and 100 in the non-background dataset. So that the feature extraction process by LBP can be analyzed using the MLP parameter approach in the form of variations in the Learning_rate network and several solvers. In the case study conducted as the methodology applied starting from data collection, algorithm development, to validation and measurement using ROC, it was found that the results of the study using the LBP method and MLP classification showed that the best accuracy results in testing the background dataset using the learning_rate network 10(-4) with Solver lbfgs were 86.66% and the best accuracy in testing the non-background dataset using the learning_rate network 10(-3) with Solver adam was 80.00%.