SINTESA CITRA DAUN KOPI MENGGUNAKAN GENERATIVE ADVERSARIAL NETWORK PADA DATASET PENYAKIT DAUN KOPI

Abstract

Coffee, as the second most traded commodity after petroleum, faces production challenges, especially due to pest or disease attacks on coffee leaves. Therefore, it is important to carry out early detection of the disease in order to minimize the risk and apply special treatment. Automatic detection of disease can be done through the application of Computer Vision technology. However, one of the main challenges faced is the limited training dataset. Generative Adversarial Networks (GANs) is a Deep Learning method that is capable of modifying images with high quality. This research aims to synthesize coffee leaf images based on the public Coffee Leaf Disease dataset using the GANs method. Testing was carried out using the RMSProp optimizer, the learning rate was 0.0001 and was carried out for 300 epochs. The architecture built uses 26 layers in the generator model and 15 layers in the discriminator model. The results of the test show that the drilled network obtained an MMSE value of 0.1658, which is not too high because the resulting synthesized image is not very good.