Multi-disease Classification and Severity Estimation of Cotton and Soybean Plants Using DenseNet

As agriculture plays vital role in nation’s economy, early diagnosis of plant diseases is a crucial and challenging task. Soybean and Cotton are the cash crops in Maharashtra therefore this study focuses on these crops. Automatic disease recognition and classification poses a variety of challenges in consideration of available datasets, tools, and image capturing conditions and has received considerable attention in the past few decades. Diseases have proven to be the root cause to major losses in the production of crop and low-quality yield. Within this scenario, the automatic disease recognition and classification is very critical and primary challenge for sustainable farming. Traditional methods were manual which are prone to errors, time-consuming and costly. In recent years Deep learning along with image processing has garnered tremendous success in a variety of application domains including automatic disease detection, but, traditional methods were not focusing on multiple diseases available on single leaf image. In this investigation, image processing techniques are investigated for their potential application in identifying cotton and soybean leaf diseases. The study evaluates the accuracy and loss of two approaches, Inception-Visual Geometry Group Network (INC-VGGN) and FACED, following various training epochs. FACED shows superior performance than INC-VGGN for both cotton and soybean plants in terms of training and validation accuracy. FACED falls behind INC-VGGN in training accuracy at first, but at the end of 30 epochs, it has caught up and even surpassed it for soybean. FACAD gives high precision and recall over INC-VGGN.