A Deep Learning Multi-Input Hybrid Model for Efficient Plant Leaf Disease Recognition

Abstract

Plant leaf disease recognition plays a very important role in smart agriculture. Advanced machine learning approaches such as deep learning have been developed and contribute to the advancement of artificial intelligence in the area for plant leaf disease recognition. Conventional approaches demand a long time and expertise. However, the implementation of new techniques continuously faces new challenges. Research into deep learning looks promising for significantly enhancing accuracy but most of the work revolves around single input image recognition to identify the plant disease. We believe that providing another kind of input like text could enhance the classification model’s effecieny. In this paper, we are using rice plant and tomato plant samples for recognition with multiple inputs and proposed a hybrid plant leaf disease recognition model that is trained using images and their text description. First, preprocess the image and descriptive text data then feature extraction of image and text descriptive information separately from two different deep learning models. Second, concatenate the image and text features. After concatenating both features, then train a classifier with new concatenated model. The model that we proposed was able to correctly identify the relevant disease with an accuracy of 99.5%.