Comparative Analysis of CNN Architectures for Maize Crop Disease

Abstract

In the current scenario, the Agriculture industry is regarded as the leading industry for society, serving all the needs for the betterment of humanity. Plants are considered to be one of the primary sources of humanity's energy production concealed with nutrients, medicinal cures, etc. Any harm or disease due to exposure of pathogens, viruses, bacteria, etc. to the plants during agriculture leads to the downfall of productivity making it a crucial concern to prevent such diseases and take necessary steps to avoid them. Making accurate identification of such fatal diseases is an important step for the industry as well as for the farmers. In our study, we have implemented fifteen different Convolutional Neural Networks (CNN) which takes plant leaf image as an input source. These architectures have different layers, neurons per layer, optimizers, etc. Our goal is to provide a detailed comparative analysis between the various frameworks based on accuracy, precision, Least Validation Cross-Entropy Loss (LVCEL), etc. parameters in the most effective way. From the experimental results, we found the sixth architecture to be the most accurate model. With the modification of convolutional layers and the use of the correct optimizer, results can be improved to a great extends.