DESIGNING DISEASE DETECTION MODEL USING CNN APPROACH FOR CORN CROP PLANTS AND COMPARING IT WITH EXISTING PRE-TRAINED MODELS

Abstract

From the beginning of human civilizations agriculture is the main occupation of human beings dealing with various factors associated with agriculture. The main objective of the agriculture is to cultivate the crops to fulfill the human requirement in different season. These crops not only fulfill the need of humans but also feed their domestic animals. Agriculture helps in growing plants for these crops that produces various grains and vegetables to fed humans but also produces crops like cotton and Jute that produces fiber which was used in manufacturing clothes wear by humans. Quantity and quality is associated with agriculture produce as it was observed by the humans which helps them in deciding fertility of agriculture land, which crops generates more yields in which season and deciding the seeds for crop. The quality of agriculture produces were directly linked to its economic value and quantity is directly linked to the prosperity of the farmers, the person who are associated with agriculture. Each crop suffers from various diseases that deteriorate both the quality and quantity of the crop yields. These diseases were caused due to various micro-organisms like bacteria and fungus, causes loses to the farmers as they affect the crop produces.

To reduce or eliminate the effect of diseases from their crops and to maintain both quantity and quality of the crop yield, farmer uses various techniques. These techniques include use of ashes of Cowdung and neem tree (Azadirachta Indica) to the various chemicals like DDT, gammaxene and pesticides etc. But the excessive use of chemicals not only disturbs the ecological balance and extinction of several species of both flora and fauna. Also now pesticides are no longer effective on new generation of diseases. These diseases now develop immunity towards the pesticides used by the farmer. Now days farmer may use various techniques and tools developed by Information Technology and may avoid lose caused to them by various crop disease. These tools may includes the various machine learning models that have the ability to find the crop disease in its beginning and precise use of specific pesticides to control the disease. These models are developed by researchers for specific crop and for some specific area. We propose a model for Maize crop and named it as M-net. We compare it with other existing models. In last part of the paper, we generalize the M-net model so that it could be used in other similar crops. We here discuss the use of Transfer learning methods.