Maize leaf disease classification using CBAM and lightweight Autoencoder network.

Maize is a widely cultivated crop in the world, so it is of great significance to identify maize diseases efficiently. We propose to introduce the Convolutional Block Attention Module (CBAM) into the autoencoder. Enhancing feature interpretability using image reconstruction techniques in maize leaf disease identification. In this work, the discrete wavelet transform was used to carry out preliminary dimensionality reduction for images. CBAM deduced the attention diagram of feature maps from space and channel respectively in the encoder, extracted image features, and eventually compressed them to Latent-space for classification. The decoder uses Latent-space as input to reconstruct the original image. In the testing phase, only the encoder is used for prediction. The experimental dataset used maize data from PlantVillage. The average accuracy of the model reached 99.44% and the training time was 123s. Compared with MobileNet, VGG16, Inception V3, ResNet-50, DenseNet201, this model has a good performance on maize data. [ABSTRACT FROM AUTHOR]