Hybrid convolutional neural network and multilayer perceptron vision transformer model for wheat species classification task: E-ResMLP+

Abstract

Wheat plant is one of the most basic food sources for the whole world. There are many species of wheat that differ according to the conditions of the region where they are grown. In this context, wheat species can exhibit different characteristics. Issues such as resistance to geographical conditions and productivity are at the forefront in this plant as in all other plants. The wheat species should be correctly distinguished for correct agricultural practice. In this study, a hybrid model based on the Vision Transformer (VT) approach and the Convolutional Neural Network (CNN) model was developed to classify wheat species. For this purpose, ResMLP architecture was modified and the EfficientNetV2b0 model was fine-tuned and improved. A hybrid transformer model has been developed by combining these two methods. As a result of the experiments, the overall accuracy performance has been determined as 98.33%. The potential power of the proposed method for computer-aided agricultural analysis systems is demonstrated.