CM-MLP: hybrid convmixer-deep MLP architecture for enhanced identification of corn and apple leaf diseases

This study aims to identify diseases impacting the agricultural sector, specifically focusing on corn and apple leaves. We propose a novel hybrid ConvMixer and Deep MLP architecture called CM-MLP that combines the strengths of ConvMixer with various Deep Multi-layer Perceptron (MLP), like MLP-Mixer, gMLP, and FNet, to tackle the classification challenges in this context. Notably, the final block of the Deep MLP can accommodate both MLP-Mixer and gMLP, as well as other designs. The ConvMixer architecture employs depthwise and pointwise convolution techniques to enrich the input image features, which are subsequently processed by the Deep MLP block to classify corn and apple leaf diseases comprehensively. In this study, we implement gradient centralization (GC) to enhance training performance. The results reveal that training with GC leads to a robust CM-MLP model, which achieves an impressive average accuracy exceeding 99.00 and proves effective in classifying diseases in corn and apple leaves.