Classification of the Severity of Downy Mildew Disease Using LGBM

Downy mildew disease attacks the leaves of melommed plants. Downy mildew spreads quickly on the surface of melon leaves. Prevention of the spread of downy mildew requires treatment according to the level of spread of downy mildew. Therefore, it is important to determine the level of disease damage on the leaf surface so that farmers' treatment of the development of downy mildew is in accordance with the level of downy mildew disease. The aim of this study was to classify the level of downy mildew damage to melon leaves using a Light Gradient Boost (LGBM). There are two concepts in the LGBM classifier algorithm: gradient-based one-side sampling (GOSS) and exclusive feature bundling (EFB). LGBM is very fast in training data, efficient, and uses low memory; therefore, it is suitable for use on low resources. This research requires a dataset of melon leaf images that are divided into four classes: the healthy leaf class, leaves affected by level 1 downy mildew disease, level 2 downy mildew disease, and level 2 downy mildew disease. The melon leaf image dataset was then preprocessed by cropping and changing the size of the melon leaf image to 128 pixels × 128 pixels. Next, the melon leaf image dataset was subjected to the process of extracting the color, texture, entropy, and edge features. The results of feature extraction, totaling 838 features, were then classified using the LGBM algorithm and tested using several parameters, including accuracy, CPU usage, memory usage, and computing time. To carry out the test, a dataset of 1861 melon leaf images was used. The test results showed an accuracy of 86.07, CPU usage of 26, 20%, 7.27Gb, and computing time of 0.44 seconds.