Implement Deep Learning Networks with Transfer Learning to Develop Energy-friendly Applications Supporting Sustainability on Image-based Plant Disease Classification

Abstract

Food security is always one of the most important factors in human lives, and crop diseases are one of the major threats which may bring potential damage. Nowadays, with the proliferation of smartphones and the advancement of machine learning methods, it is more likely to achieve rapid identification of disease diagnosis by a smartphone-assisted application supported by deep learning trained models. By comparing different datasets and different kinds of CNN frameworks, this paper trained deep convolutional neural networks based on plant leaves’ images to identify species and detect diseases. Furthermore, this paper found the best combination of different datasets with the highest accuracy. The highest accuracy this work got is 97.37%, using ResNet-9 along with Transfer Learning. Nevertheless, these training datasets are too straightforward to deal with the more complex real-world situation. Besides, two-dimensional datasets from time to time have such limited information; therefore, more information is needed to diagnose plants’ diseases. For future extension, this work can apply not only image datasets but also environmental factors, such as soil structure and image background, to construct a more precise model to diagnose plant diseases. Hence, the concept of Point Cloud will be discussed in this paper. This work can be viewed as the first step to build an Energy-friendly plant disease classification application supporting sustainability.