The analysis of the implementation of convolutional neural network architectures for coffee leaf disease image classification.

Coffee is an important international commodity, especially for Indonesia, which ranks fourth as the world's coffee producer. However, in the field, coffee plants are susceptible to pests and diseases, which can reduce the yield and quality of the finished product. For farmers who often rely on rain-fed crops and have limited access to financial and technical support, this is a significant obstacle. Image processing and machine learning techniques, particularly deep learning and convolutional neural networks (CNN), can help identify and early detect plant diseases. This study used a quadcopter drone for image acquisition and employed CNN architectures, specifically AlexNet with 50, 75, and 100 epochs, to classify coffee leaves into three classes: healthy, leaf rust, and leaf spot. The best results were achieved by Alexnet with 100 epochs with an accuracy of 93.7%, precision of 98%, and recall of 100%. Thus, it can be concluded that the CNN method using the Alexnet model with 100 epochs and combined with a quadcopter drone can be used to detect coffee plant diseases. [ABSTRACT FROM AUTHOR]