EfficientNet-Based Sugarcane Disease Classification with Dual-Convolution Spatial Attention CBAM (EfficientNet-DCCBAM)

Agriculture has long faced challenges from harmful insect pests, including those affecting sugarcane, a vital crop in Indonesia. Traditional control methods such as insecticides and pest monitoring techniques have been employed to mitigate the impact of these pests. In the context of precision agriculture, artificial intelligence (AI) technology enables the efficient detection and classification of organisms. One AI technique, deep learning, has been widely applied in the classification of plant diseases. This research focuses on the application of deep learning for the classification of sugarcane diseases. We propose a deep learning network architecture that integrates EfficientNetV2B0 with Dual Convolution Spatial Attention CBAM (DCSA-CBAM) to enhance the features of infected sugarcane leaf images. This paper introduces a new dataset comprising 1968 sugarcane leaf images from various regions in East Java. Our proposed model achieved an accuracy of $\mathbf{9 4 . 2 \%}$, demonstrating strong performance in classifying sugarcane diseases. We also compare the performance of several other deep learning networks, including DenseNet, MobileNet, ResNet, and VGG with our model showing superior results.