For crops to be produced, plant diseases must be identified as soon as possible. Numerous outstanding, cutting-edge deep learning methods have been developed for this topic. Conventional approaches to disease identification and classification frequently depend on large-scale labelled datasets, which can be expensive and time-consuming to obtain. Few-shot learning trains models that are highly generalizable to fresh, unseen data by using a limited number of instances per class. Our method uses little labelled data to enable accurate and efficient plant disease identification by combining the power of deep neural networks and transfer learning. In this paper we have analysed the traditional method CNN, on plant disease detection it will give promising results using that as a base network FSL will analysed with different approaches like Few shot with Siamese Network, Few shot with Prototypical network and Few shot with triplet loss function, Plant village data set with divide in three different settings with different N way K shot, N is taken as 5 and Shot will take as 1, 5, 10, 20, 30, 40, 50, etc. As the value of shot increases it increases the accuracy also. On the 5 way 50 shot we get maximum accuracy which is 95.27% with the limited sample sit giving a better performance.