A Comparison of CNN Architectures Using Transfer Learning to Detect Paddy Leaf Diseases

Abstract

I provide a method for real-time detection of Paddy Leaf Disease in my article. Blast, brown spot, and bacterial leaf blight are among the most prevalent diseases that may harm rice plants in my nation. Farmers may be losing money due to decreased agricultural yields caused by harmful pathogens. The phrase "paddy leaf disease" describes a group of illnesses that impact rice plants. These diseases may be caused by fungi, bacteria, or viruses and because the leaves to become yellow, wilt, and eventually reduce the crop's output. For this reason, I need to find that issue quickly so that I may harvest additional crops. To enhance crop health, boost yields, and decrease economic losses, farmers may use deep learning to detect paddy leaf disease. This approach is quicker, more accurate, and more consistent than traditional methods of disease identification and diagnosis in rice plants. Discovering trends and patterns in illness data may also be aided by this, which can result in fresh perspectives and advancements in the agricultural sector. I need a procedure, such as data collecting, in order to complete the process or create that Deep-learning. From the internet, I get a dataset of images. The collection contains five different kinds of images. I proceed to pre-process the dataset. The next step is to train my model and put it through its paces in testing. Lastly, I use a number of algorithms, including DenseNet121, ResNet50V2, MobileNetV2, and EfficientNetB2. Using EfficientNetB2, I was able to get an accuracy of about 91%. The techniques used in each person's comparison statements are also included in the article's implementation component. To build the most accurate model for the conditions, this study also makes use of model validation techniques.