Plant Disease Detection Using Optimised Cnn Models for Resource Constrained Environment

Agricultural productivity is helpless against various diseases that result in significant losses in terms of resources. These losses rise rapidly in isolated locations with limited resources. The total annual harvest losses resulting from plant diseases are estimated to be close to $ 60 billion. Image recognition can aid in the quicker and more accurate analysis of plant diseases since these conditions are evaluated based on their visual signs. One common use of deep learning models based on convolutional neural networks (CNNs) is the detection of plant diseases. They must, however, be optimised for devices with limited resources, such as smartphones. Along with the training dataset's class im-balancing issue, this is a serious issue. To address this issue, we employ transfer learning and data augmentation techniques to improve the model's accuracy for the target dataset. We make use of the 55,000-entry Plant Village Dataset. Nowadays, cell phones are affordable, and when combined with the internet's limitless capabilities, they can be a suitable platform for plant disease analysis. Because the Mobile-Net CNN model is lightweight and delivers excellent accuracy with transfer learning, we want to employ it. Thus, this application will be very helpful for the farmers, especially those in remote areas, which have even fewer resources to cut their yield misfortunes from diseases, as they will be able to tackle these problems with the ability to detect the diseases and perform immediate remedies.