Recognition and Classification of Maize Drought on UAV Hyperspectral Images
author: Gavin
2022-01-06
Traditional crop drought identification methods mostly use satellite multi-spectral remote sensing images to calculate vegetation index to obtain drought conditions. However, due to the limitations of weather, transit time, low spatial resolution and other factors, satellite images are not intimidating in terms of timeliness and accuracy. satisfy. With the development of unmanned aerial vehicle technology, it has become possible to obtain ultra-high-resolution remote sensing images with convenient operation, flexibility and effectiveness. In recent years, it has provided new ideas for solving many agricultural problems. With the development of deep neural networks, the proposal of fully convolutional neural networks has further improved the accuracy of remote sensing image semantic segmentation tasks.
Principle
The cause of agricultural drought is that the soil moisture content at the roots of the plants is reduced, and the plants cannot absorb enough water molecules to grow. Severe water loss will reduce the transpiration of crops, and the stomata on the leaves will shrink or even close, causing crops to wither. Therefore, both surface temperature and vegetation index can be used to monitor the drought of crops.Vegetation information is mainly reflected in red (R) and near infrared spectroscopy (Near Infrared, NIR). The normalized difference vegetation index (NDVI) calculated through these two bands can effectively feed back the growth status of vegetation.
Solution
The data in this study is a four-channel (RGB+NIR) multispectral image taken by the DJI Jingwei-M600-PRO drone. The drone is equipped with a hyperspectral imager ATH9010 produced by Optosky Photonics.Use the NDVI values calculated in the red (R) and near-infrared (NIR) bands to make labels, and train on the drone RGB image samples through the semantic segmentation method based on deep learning, so that the trained network can be used in ordinary consumer-level unmanned The RGB image taken by the machine is used to identify and classify the corn drought.
The overall process of the solution is as follows:
(1) Selection of original data: This paper selects corn image data in villages and towns affected by drought in early autumn for research.
(2) Perform preprocessing such as screening and splicing of data; divide training set, validation set and test set; crop the training set and validation set samples, and perform edge mirroring and area overlap cropping on the test set samples.
(3) Manually divide the non-maize planting areas such as roads and houses, that is, the "other" category; use the red (R) and near-infrared (NIR) bands to calculate the NDVI index of the corn area and perform smoothing processing, combined with expert evaluation for corn Evaluation of the drought level, generating a disaster level label.
(4) Build a network model.
(5) Input the training set and validation set samples into the network for training, and set reasonable training parameters.
(6) Test the trained network on the test, and use evaluation indicators such as Fl-score>Jaccard coefficient to evaluate the accuracy of the network.
(7) Optimize the structure and parameters of the network, and repeat steps (4) to (6) for comparison experiments, until the test set scores meet the requirements and no longer increase significantly, and the optimal network model is obtained.
(8) Visualize test results.
Test result
Principle
The cause of agricultural drought is that the soil moisture content at the roots of the plants is reduced, and the plants cannot absorb enough water molecules to grow. Severe water loss will reduce the transpiration of crops, and the stomata on the leaves will shrink or even close, causing crops to wither. Therefore, both surface temperature and vegetation index can be used to monitor the drought of crops.Vegetation information is mainly reflected in red (R) and near infrared spectroscopy (Near Infrared, NIR). The normalized difference vegetation index (NDVI) calculated through these two bands can effectively feed back the growth status of vegetation.
Solution
The data in this study is a four-channel (RGB+NIR) multispectral image taken by the DJI Jingwei-M600-PRO drone. The drone is equipped with a hyperspectral imager ATH9010 produced by Optosky Photonics.Use the NDVI values calculated in the red (R) and near-infrared (NIR) bands to make labels, and train on the drone RGB image samples through the semantic segmentation method based on deep learning, so that the trained network can be used in ordinary consumer-level unmanned The RGB image taken by the machine is used to identify and classify the corn drought.
The overall process of the solution is as follows:
(1) Selection of original data: This paper selects corn image data in villages and towns affected by drought in early autumn for research.
(2) Perform preprocessing such as screening and splicing of data; divide training set, validation set and test set; crop the training set and validation set samples, and perform edge mirroring and area overlap cropping on the test set samples.
(3) Manually divide the non-maize planting areas such as roads and houses, that is, the "other" category; use the red (R) and near-infrared (NIR) bands to calculate the NDVI index of the corn area and perform smoothing processing, combined with expert evaluation for corn Evaluation of the drought level, generating a disaster level label.
(4) Build a network model.
(5) Input the training set and validation set samples into the network for training, and set reasonable training parameters.
(6) Test the trained network on the test, and use evaluation indicators such as Fl-score>Jaccard coefficient to evaluate the accuracy of the network.
(7) Optimize the structure and parameters of the network, and repeat steps (4) to (6) for comparison experiments, until the test set scores meet the requirements and no longer increase significantly, and the optimal network model is obtained.
(8) Visualize test results.
Test result
| NO | NDVI | Extent of disaster |
| 0 | 0.5 〜1 | Unaffected |
| 1 | 0.42 〜0.5 | Mild drought |
| 2 | 0.36 〜0.42 | Moderate drought |
| 3 | 0 〜0.36 | Severe drought |
口Unaffected ■Mild drought 口Moderate drought ■Severe drought ■other
Before slicing the test image, copy 50 pixels of each edge as mirror filling. The reasoning process uses overlapping sliding window slices as input, the edges overlap by 100 pixels, and crop the 512x512 image input network . The output size of the network inference is also 512X512, discarding the 50 pixels of the edge, leaving the size of the middle area as 412X412, and splicing the results together.
Conclusion
In this article, we focus on how to use the RGB images obtained by drones to identify corn droughts and classify droughts through deep learning semantic segmentation. First, by analyzing the four-channel UAV data, combined with expert daily interpretation to generate a maize crop disaster level distribution map. Secondly, according to the existing tags, combined with the semantic segmentation network of three-channel input, the disaster classification of deep learning is realized, which proves the feasibility of deep learning in disaster analysis. Finally, by modifying the loss function, optimizing the U-Net structure and other operations, the performance of disaster area recognition is improved.
Related products
Vegetation information is mainly reflected in red (R) and near infrared spectroscopy (Near Infrared, NIR)
Hyperspectral Camera:ATH1010
Airborne Hyperspectral Remote Sensing System:ATH9010
Drone Hyperspectral Imaging System:ATHL9010
Related articles
Conclusion
In this article, we focus on how to use the RGB images obtained by drones to identify corn droughts and classify droughts through deep learning semantic segmentation. First, by analyzing the four-channel UAV data, combined with expert daily interpretation to generate a maize crop disaster level distribution map. Secondly, according to the existing tags, combined with the semantic segmentation network of three-channel input, the disaster classification of deep learning is realized, which proves the feasibility of deep learning in disaster analysis. Finally, by modifying the loss function, optimizing the U-Net structure and other operations, the performance of disaster area recognition is improved.
Related products
Vegetation information is mainly reflected in red (R) and near infrared spectroscopy (Near Infrared, NIR)
Hyperspectral Camera:ATH1010
Airborne Hyperspectral Remote Sensing System:ATH9010
Drone Hyperspectral Imaging System:ATHL9010
Related articles
- Shiferaw B, Prasanna B M, Hellin J, et al. Crops that feed the world 6. Past successes and future challenges to the role played by maize in global food security [J]. Food Security, 2011, 3(3): 307.
- Campos H, Cooper M, Habben J E, et al. Improving drought tolerance in maize: a view from industry [J]. Field Crops Research, 2004, 90(1):0-34.
- Wu T, Zhang L, Peng B, et al. Real-time progressive hyperspectral remote sensing detection methods fbr crop pest and diseases[C]// SPIE Commercial + Scientific Sensing and Imaging. International Society for Optics and Photonics, 2016.
- Howard J A, Barrett E C, Heilkema J U. The application of satellite remote sensing to monitoring of agricultural disasters [J]. Disasters, 1978, 2(4):231-240.
- Xiang T Z, Xia G S, Zhang L. Mini-Unmanned Aerial Vehicle-Based Remote Sensing: Techniques, applications, and prospects [J]. IEEE Geoscience and Remote Sensing Magazine, 2019, 7(3): 29-63.
- Linchant J, Lisein J, Semeki J, et al. Are unmanned aircraft systems (UASs) the future of wildlife monitoring? A review of accomplishments and challenges [J]. Mammal Review, 2015, 45(4):239-252.
Reseach on Classification of Pinus Massoniana Pests Based on Hyperspectral Data
Monitoring of River oil Spill Using UAV Hyperspectral Remote Sensing
Related Article
Discover how UAV‑based hyperspectral imaging and deep learning can monitor SDI, TOC, and TEP – key membrane fouling indicators for seawater desalination during harmful algal blooms. Based on a recent Water Research study, this article explores how Optosky's ATH9010 enables spatial risk mapping and proactive intake management.
Applications | ATH9010 Enables Water Quality and Membrane Fouling Monitoring for Desalination During Algal Blooms
Choosing between 785 nm and 1064 nm for Raman? This guide explains the physics of fluorescence, compares signal strength and fluorescence suppression, and provides a step‑by‑step decision tree. Real‑world drug detection and pesticide examples show why wavelength matters – and how dual‑wavelength coverage offers the ultimate solution.
785 nm or 1064 nm? Fluorescence Is the Dividing Line
Inspired by the 2026 Science paper from Zhang Jun’s team on a video‑rate on‑chip hyperspectral microsystem, Optosky’s ATH series moves beyond lab‑only tools. We deliver tailored, real‑world hyperspectral systems that integrate compact hardware, edge AI, and application‑specific algorithms—turning raw spectral data into actionable decisions, right where you need them.
From Science to the Field: Custom Hyperspectral Solutions by Optosky
See how 40 Optosky NY3300Pro multispectral units were deployed across a large‑scale demonstration farmland in Northwest China. With 9‑band imaging, solar power, 4G transmission, and automated data analytics, this solution enables precise growth monitoring, early pest detection, and water‑fertiliser optimisation – a true leap toward smart agriculture.
Case Study | Bulk NY3300Pro Installation Transforms Smart Farming