Citation format:Jianing Long, Zhao Zhang, Xiaohang Liu, Yunxia Li, Zhaoyu Rui, Jiangfan Yu, Man Zhang, FLORES Paulo, Zhexiong Han, Can Hu, Xufeng Wang. Detection of Lodging Types of Wheat Using Improved EfficientNetV2 and Unmanned Aerial Vehicle (UAV) Images[J]. Smart Agriculture, 2023, 5(3): 62-74.
DOI:10.12133/j.smartag.SA202308010
LONG Jianing, ZHANG Zhao, LIU Xiaohang, LI Yunxia, RUI Zhaoyu, YU Jiangfan, ZHANG Man, FLORES Paulo, HAN Zhexiong, HU Can, WANG Xufeng. Wheat lodging types detection based on UAV image using improved EfficientNetV2[J]. Smart Agriculture, 2023, 5(3): 62-74.
DOI:10.12133/j.smartag.SA202308010
The full text on the official website is free to read
CNKI Reading
Improved EfficientNetV2 and UAV images were used to detect lodging types in wheat
Long Jianing1,2, Zhang Zhao1,2*, Liu Xiaohang1,2, Li Yunxia1,2, Rui Zhaoyu1,2, Yu Jiangfan1,2, Zhang Man1,2, FLORES Paulo3, Han Zhexiong4,5, Hu Can6, Wang Xufeng6
(1.College of Information and Electrical Engineering, China Agricultural University, Beijing 100080, China; 2. Key Laboratory of Agricultural Information Access Technology, Ministry of Agriculture and Rural Affairs, China Agricultural University, Beijing 100083, China; 3. Department of Agricultural and Biological Engineering, North Dakota State University, Fargo, NC 58102, USA; 4. Department of Biosystems Engineering, Gangwon National University, Chuncheon, Gangwon 24341, Korea; 5. Interdisciplinary Department of Smart Agriculture, Gangwon National University, Chuncheon, Gangwon 24341, South Korea; 6. School of Mechanical and Electrification Engineering, Tarim University, Alar 843300, Xinjiang, China)
Summary:
Different types of wheat lodging (root lodging, stem lodging) have different effects on yield and quality. The purpose of this study was to classify the lodging types of wheat by UAV images and explore the effect of UAV flight altitude on classification performance.
[Method] Three UAV flight altitudes (15, 45 and 91 m) were set up to obtain the images of wheat experimental fields, and the automatic segmentation algorithm was used to generate datasets at different altitudes, and an improved EfficientNetV2-C model was proposed to classify and identify them. The model introduces the Coordinate Attention (CA) attention mechanism to improve the network feature extraction ability, and combines CB-Focal Loss (Class–Balanced Focal Loss) to solve the impact of data imbalance on the classification accuracy of the model.
The improved EfficientNetV2-C performed the best, with an average accuracy of 93.58%. Compared with the four unimproved machine learning classification models (Support Vector Machine (SVM), K Nearest Neighbor (KNN), Decision Tree (DT) and Naive Bayes (NB)) and two deep learning classification models (ResNet101 and EfficientNetV2), Among them, EfficientNetV2 has the best performance at all altitudes, with an average accuracy of 82.67%. The flight altitude of the UAV has no significant effect on the performance of the four machine learning classifiers, but the classification performance of the deep learning model decreases due to the loss of image feature information as the flight altitude increases.
[Conclusion] The improved EfficientNetV2-C achieves high accuracy in the detection of wheat lodging type, which provides a new solution for wheat lodging early warning and crop management.
Key words: lodging type of wheat; image processing; deep learning; unbalanced data; Machine learning; drone
Image of the article
Fig.1 Flow chart of wheat lodging type detection
Fig. 1 Flowchart of wheat lodging types detection
Figure 2 The process diagram is automatically generated from the model dataset
Fig. 2 Diagram of the process of automatic model dataset generation
Fig.3 Schematic diagram of UAV images of different types of wheat lodging
Fig. 3 Schematic diagram of different wheat lodging types based on UAV images
Figure 4 Visualization of data augmentation mode
Fig. 4 Visualization of the data enhancement approach
Figure 5 EfficientNetV2-C network structure
Fig. 5 EfficientNetV2-C network structure diagram
Figure 6 Schematic diagram of attention mechanism
Fig. 6 Schematic diagram of the attention mechanism
Fig.7 Classification results of four machine learning classifiers (SVM, DT, KNN, NB).
Fig. 7 Classification results of four machine learning classifiers （SVM， DT， KNN， NB）
Fig.8 The training process of improving the EfficientNetV2-C model to train and verify the accuracy and loss of the model at different altitudes
Fig. 8 Improvement of EfficientNetV2-C model for training and validation of model accuracy and loss training
process at different altitudes
Note: The left side of each small plot shows the prediction results of the long rectangular test field with a size of 1.5 m × 15 m. The prediction results of the short rectangular test field with a size of 1.5 m × 3.7 m are shown on the right
Fig.9 Prediction of three types of wheat lodging at different heights by EfficientNetV2-C
Fig. 9 EfficientNetV2-C prediction of three types of collapse at different heights
About the corresponding author
Prof. Zhao Zhang
Zhang Zhao, professor and doctoral supervisor of China Agricultural University, "Outstanding Talent Introduction" of China Agricultural University, part-time researcher of North Dakota State University, editor-in-chief of Springer's smart agriculture book series, and youth editorial board member of IJABE, ASABE, "Smart Agriculture (Chinese and English)" and other journals, the main research direction is the intelligent perception of crop phenotype and the research and development of smart orchard operation equipment. The relevant results have been published in journals such as Transactions ASABE, Applied Engineering in Agriculture, Computers and Electronics in Agriculture, and Postharvest Biology and Technology, and more than 30 SCI/EI academic papers have been published as the first or corresponding author, and 1 invention patent has been authorized. He has published 3 English monographs, presided over 6 precision agriculture projects in the past 3 years, and participated in 3 other projects.
Source: Smart Agriculture, Issue 3, 2023
Please contact the editorial department for authorization for reprinting
Supported units in this issue
Weichai Lovol Smart Agricultural Technology Co., Ltd
Zibo Institute of Digital Agriculture and Rural Studies
Shanghai Zanqi Culture Technology Co., Ltd
Smart agriculture WeChat exchange service group
In order to facilitate the academic exchange of readers, authors and reviewers in the field of agricultural science, promote the development of smart agriculture, and better serve the readers, authors and reviewers, the editorial department has established a WeChat communication service group, where questions related to issues in the professional field and submissions can be consulted. How to join the group: Add the editor's WeChat 331760296, note: name, unit, research direction, the editor pulls you into the group, and the institutional marketing and advertising personnel do not disturb.
Post a call for submissions
Welcome to publish information such as the introduction of the scientific research team, innovative scientific research achievements and related activities on our official account.