GrapeNet: A Lightweight Convolutional Neural Network Model for Identification of Grape Leaf Diseases

Most convolutional neural network (CNN) models have various difficulties in identifying crop diseases owing to morphological and physiological changes tissues, cells. Furthermore, a single disease can show different symptoms. Usually, the differences symptoms between early late stages include area of color disease. This also poses additional for CNN models. Here, we propose lightweight model called GrapeNet identification symptom specific grape diseases. The main components are residual blocks, feature fusion blocks (RFFBs), convolution block attention modules. used deepen depth extract rich features. To alleviate performance degradation associated with large number hidden layers, designed an RFFB module based on block. It fuses average pooled map before input high-dimensional maps after output by concatenation operation, thereby achieving at depths. In addition, (CBAM) is introduced each valid information. obtained results that accuracy was determined as 82.99%, 84.01%, 82.74%, 84.77%, 80.96%, 83.76%, 86.29% GoogLeNet, Vgg16, ResNet34, DenseNet121, MobileNetV2, MobileNetV3_large, ShuffleNetV2_×1.0, EfficientNetV2_s, GrapeNet. achieved best classification when compared other classical total parameters only included 2.15 million. Compared which has highest among models, reduced 4.81 million, reducing training time about two times DenseNet121. Moreover, visualization Grad-cam indicate introduction CBAM emphasize information suppress irrelevant overall suggest useful automatic leaf