<![CDATA[The decline in rice productivity in Indonesia is often caused by drought and leaf diseases that are difficult to detect early. This condition requires a technology-based classification system that is able to provide fast and accurate diagnosis as support for decision making in the agricultural sector. This study aims to develop a rice leaf image classification model using the CSPDarknet architecture, with a color and texture feature extraction approach. The dataset used is the result of primary documentation that has gone through an augmentation process to increase the diversity of training data. The model architecture consists of a CSPDarknet backbone combined with a Cross-stage Partial Bottleneck with two Convolutions (C2f) block, Spatial Pyramid Pooling - Fast (SPPF), Global Average Pooling, and dropout to improve model generalization. Training was carried out using the Stratified 5-Fold Cross-Validation method and three optimizer variations, namely Stochastic Gradient Descent (SGD), Adam, and AdamW. The experimental results showed that the best model combination was achieved with the AdamW optimizer, with an average accuracy value of 99.72%, precision of 99.73%, recall of 99.72%, and F1-score of 99.72%. These findings indicate that the proposed classification approach is able to effectively distinguish healthy, diseased, and drought-affected leaves. In the future, this model has the potential to be further developed through the integration of Raspberry Pi-based Internet of Things (IoT) devices for real-time monitoring of plant conditions in the field.]]>
