Loading…

Identifying plant diseases using deep transfer learning and enhanced lightweight network

Plant diseases can cause significant reductions in both the quality and quantity of agricultural products, and they have a disastrous impact on the safety of food production. In severe cases, plant diseases may even lead to no grain harvest completely. Therefore, seeking fast, automatic, less expens...

Full description

Saved in:
Bibliographic Details
Published in:Multimedia tools and applications 2020-11, Vol.79 (41-42), p.31497-31515
Main Authors: Chen, Junde, Zhang, Defu, Nanehkaran, Y. A.
Format: Article
Language:English
Subjects:
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Plant diseases can cause significant reductions in both the quality and quantity of agricultural products, and they have a disastrous impact on the safety of food production. In severe cases, plant diseases may even lead to no grain harvest completely. Therefore, seeking fast, automatic, less expensive and accurate methods to detect plant diseases is of great realistic significance. In this paper, we studied the transfer learning for the deep CNNs and modified the network structure to enhance the learning ability of the tiny lesion symptoms. The pre-trained MobileNet-V2 was extended with the classification activation map (CAM), which was used for visualization as well as plant lesion positioning, and both were selected in our approach. Particularly, the transfer learning was performed twice in model training: the first phase only inferred the weights from scratch for new extended layers while the bottom convolution layers were frozen with the parameters trained from ImageNet; the second phase retrained the weights using the target dataset by loading the model trained in the first phase. Then, the yielded optimum model was used for identifying plant diseases. Experimental results demonstrate the validity of the proposed approach. It achieves an average recognition accuracy of 99.85% on the public dataset. Even under multiple classes and complex background conditions, the average accuracy reaches 99.11% on the collected plant disease images. Thus, the proposed approach efficiently accomplished plant disease identification and presented a superior performance relative to other state-of-the-art methods.
ISSN:1380-7501
1573-7721
DOI:10.1007/s11042-020-09669-w