Title: Identifying rice leaf diseases using an improved AlexNet model
Authors: Le Yang; Xiaoyun Yu; Mingfu Liao; Shaoping Zhang; Huibin Long; Huanhuan Zhang; Yuanjun Liao
Addresses: School of Computer and Information Engineering, Jiangxi Agricultural University, Nanchang 330045, China ' School of Computer and Information Engineering, Jiangxi Agricultural University, Nanchang 330045, China ' School of Computer and Information Engineering, Jiangxi Agricultural University, Nanchang 330045, China ' School of Computer and Information Engineering, Jiangxi Agricultural University, Nanchang 330045, China ' School of Computer and Information Engineering, Jiangxi Agricultural University, Nanchang 330045, China ' School of Computer and Information Engineering, Jiangxi Agricultural University, Nanchang 330045, China ' School of Computer and Information Engineering, Jiangxi Agricultural University, Nanchang 330045, China
Abstract: In order to facilitate farmers to accurately identify rice leaf diseases, this study improves the original AlexNet model to identify eight kinds of rice leaf diseases. The original AlexNet model was improved by changing the original Relu activation function to Leakyrelu activation function, as well as removing the last convolution layer and changing the number of convolutional kernels of each original convolutional layer to 1/2 of the original. Finally, the output nodes of the full connection layer were reduced. The results show that the recognition rate of the improved model reaches 99.23%, significantly higher than that of the original, and so, a new idea for the identification of rice leaf diseases was put forward. In addition, this paper verifies the generalisation of the model by designing five comparative experiments with different activation functions, different dataset sizes, different optimisers, different batch sizes, and different learning rates.
Keywords: rice leave; disease; identification; AlexNet; deep learning.
DOI: 10.1504/IJWMC.2023.132422
International Journal of Wireless and Mobile Computing, 2023 Vol.25 No.1, pp.1 - 10
Received: 06 Jun 2022
Received in revised form: 05 Jan 2023
Accepted: 12 Jan 2023
Published online: 19 Jul 2023 *