Model Selection of Hybrid Feature Fusion for Coffee Leaf Disease Classification

Coffee leaf diseases can significantly impact the productivity and quality of crops. Accurate timely identification these is crucial for effective management control. This paper proposes a hybrid feature fusion approach identifying coffee disease, including early late fusion. First, we propose several models to extract information in input images by combining MobileNetV3, Swin Transformer, variational autoencoder (VAE). acting on inductive bias locality, image features that are closer one another (local features), while Transformer able interactions further apart (high-level features). These differently extracted contain complementary enriches unified map. Second, from fused network. The early-fusion learner network deployed learn rich feature. implemented comprehensively before classification classifies diseases. proposed evaluated challenging, real world Robusta Leaf (RoCoLe) dataset with various diseases, red spider mite rust disease. results show our approach, MobileNetV3 outperforms individual an accuracy 84.29%. In conclusion, promising disease identification, providing accurate diagnosis control real-world conditions.