An Improved Crop Disease Identification Based on the Convolutional Neural Network

Abstract

: The increase in population leads to an increase in the need for food production. A healthy, pest-free plant can providea considered amount of yield in time. However, many conditions affect crop production.Identifying crop disease accurately, fast, and intelligently, plays an important role in agriculture informatization development. Most existing methods are performed manually, which affects the identifyingresult. Careful monitoring and diagnosis of crops for a large area manually is a tedious process. To address these issues, we proposed an improved crop disease identification based on the convolutional neural network (CNN) architecture.The first operation is to resize crop images and to be normalizedthem.Here, we built a neural network toload data samples for training and dividedthe verification set and training set. Furthermore, to adjust the learning rate dynamically, we use Adam algorithms which combinedthe RMSprop algorithm and momentum algorithm to improve the training learning rate.Finally, we used PlantVillage dataset to carry out the validations, this dataset contains 38 different types of crops. The experimentalresult showed the test accuracy and validation accuracy are100% and 97.50% respectively. Compared with state-of-the-art methods, our proposed model has higher detection accuracy. The convolutional neural network proposed in this paper has a high accuracy and fast training speed. The proposed architecture is less time-consuming which can help to improve the development of smart agriculture.