Crop Type and Stress Detection using Transfer Learning with MobileNet

Image classification has been used to identify and classify different crops and tell us about their health condition. It is in high demand to investigate for an on-the-fly classification for crop discrimination and stress identification having a camera installed on the UAV. This paper proposes a new MobileNet+D3 architecture to identify the crop class from the leaf images and to detect the crop stress. In this study, the leaf images of 10 different crops have been collected from both healthy and unhealthy plants. The proposed model is compared with traditional machine learning algorithms like KNN and SVM along with a Deep CNN model with different activation functions like ReLU, Leaky ReLU and Adam optimizer. The proposed MobileNet+D3 model achieved an accuracy of 97% which is higher than all the methods considered in the comparative study. The proposed method is more suitable for real-time classification on low computing devices.