uav-based HYperspectral monitoring of aGRIculture
HYGRI aims to acquire and use UAV-based hyperspectral images to monitor and detect plant pathogens in fruit trees and to visualize the information in a user-friendly decision support system. Every object has its own spectral signature, including diseased plants. Thanks to the possibility to calculate accurate spectral indices by narrow-band combination in the visible and near-infrared ranges and the use of ML algorithms, HYGRI will detect these plants and help farmers in their daily work, by reducing the spread of disease and the need to cut plants, and will benefit the environment we live in and the food we eat by allowing treatments to be targeted only at specific plants. Existing know-how at AVT-ASI provides the basis for operating the UAV system and detecting plants affected by a number of diseases, for example flavescence dorée in vineyards, and for analysing plant health status, but advances are needed to solve challenges in the acquisition and processing phases and in the choice of the machine learning approach. In this context the main objective of HYGRI is to make the UAV flights more efficient and obtain images with better data quality, to continue the validation of ML algorithms with ground truth and to commercialize services and an easy-to-use tool for professional use. The objective is realized through Specific Objectives (SO), defined as follows:
- SO 1: to optimize the flight planning and acquisition of hyperspectral images with drones by taking into account advanced settings and parameters and illumination conditions;
- SO 2: to improve the quality of UAV-based hyperspectral images after pre-processing and automate the workflow;
- SO 3: to improve the performance of in-house ML algorithms to early identify the plants affected by the disease with rigorous validation with field measurements;
- SO 4: to evaluate the use of the solution for other applications of green analysis (urban green, forestry, other crop diseases).
- SO 5: to commercialize the monitoring services and the decision support system (DSS) internationally.
Tech components and data
AVT-ASI will use the hyperspectral sensor AFX10 by Specim that acquires 224 bands in the 400-1000 nm range. The camera is mounted on a UAV DJIM600 Pro. To guarantee the stability of the camera, a Gremsy T7 gimbal is used. Georeferencing is provided through GPS with AV14 antenna and IMU integrated in the AFX10. Flight planning is done with UgCS by SPH Engineering. After the flight the images are both radiometrically and geometrically corrected using SPECIM’s CaliGeoPRO software. Geometric correction rectifies spatial distortions caused by sensor characteristics and terrain relief, aligning the image with geographic coordinates for reliable spatial analysis and mapping. However a proper aerial triangulation is not implemented and residual errors might be present. The atmospheric correction compensates for atmospheric effects on incoming radiation, including mitigating the impacts of atmospheric water vapour, optical thickness, solar position, and topographic illumination differences, and is executed in DROACOR software, developed by the ReSe Applications LLC, that has a higher level of automation with respect to other scripts like 6S or MODTRAN-based atmospheric simulation processes. Machine learning algorithms in use in AVT-ASI are typically Random Forest, Support Vector Machine and Neural Network. Scripts are available in R and Python languages. An existing DSS solution has been implemented with Java and HTML scripts and published on a cloud server for demonstration purposes. For visualization, data manipulation and editing ENVI and QGIS software packages are used.
Challenges and how they will be addressed
In the last couple of years AVT-ASI has invested in the purchase of the hyperspectral camera AFX10 by Specim and the drone DJI M600 Pro, and in the development of the workflow application, which allows various types of analyses in different environments. The pilot projects run at AVT-ASI have proved that hyperspectral images provide technologically advanced, precise and reliable data, which allow obtaining results not obtainable with traditional technologies. However some challenges (C) have emerged that HYGRI aims to resolve:
– C1: the quality of hyperspectral UAV images is highly dependent on factors that are generally not significant with multispectral sensors, such as the sun illumination, the orientation of the crop lines with respect to the sun azimuth, the effect of shadows, the time differences between strips;
– C2: the geometric pre-processing of hyperspectral images shows uncertainties due to the fact that commercial software do not apply a rigorous approach (i.e. image triangulation) for image georeferencing;
– C3: the data processing operations are executed in a fragmented way in different software packages or scripts with high manual intervention;
– C4: the validation of the UAV-based results on early-detection plant disease, in particular flavescence dorée, has to be further investigated with more ground truth and multi-temporal comparisons;
– C5: to replicate the approach on other diseases or phenomena, an efficient methodology for ML training is not available;
– C6: the current visualization tool does not meet all the user requirements.
The challenges will be addressed by defining and following Specific Objectives (SO), such as:
SO 1: to optimize the flight planning and acquisition of hyperspectral images with drones by taking into account advanced settings and parameters and illumination conditions;
SO 2: to improve the quality of UAV-based hyperspectral images after pre-processing and automate the workflow;
SO 3: to improve the performance of in-house ML algorithms to early identify the plants affected by the disease with rigorous validation with field measurements;
SO 4: to evaluate the use of the solution for other applications of green analysis (urban green, forestry, other crop diseases).
SO 5: to commercialize the monitoring services and the decision support system (DSS) internationally.
Expected outcomes
1
Improved experience on UAV hyperspectral image acquisition
2
Improved know-how in field surveys
3
Improved accuracy of disease mapping
4
Replicability of HYGRI on another environments
5
Release of HYGRI tool
6
Commercialization of the services
About AVT Airborne Sensing Italia srl
| Full name | AVT Airborne Sensing Italia srl |
| Country | Italy |
| Website | http://www.avt-as.eu/ |
| Contact Persons | Daniela Poli, Hafiz Ali Imran |
| Contact’s email | , |
Based in Trento (Italy), AVT Airborne Sensing Italia (AVT-ASI) is subsidiary of AVT Airborne Sensing, a group with decades of experience in aerial surveying, photogrammetry and remote sensing. Our fleet includes 5 airplanes, equipped with a variety of sensors for different applications: photogrammetric cameras, LiDAR, thermal camera, hyperspectral camera. We have qualified personnel in charge of data processing (orthophotos, DSM/DTM, cartography, mapping), based in Italy, Austria and Germany.
AVT-ASI main competences are:
- UAV-based remote sensing: aerial surveys with hyperspectral camera AFX10 by Specim, image preparation, mosaicking
- Spectral analysis
- Thematic mapping with AI solutions, health analysis of crops
- Applications in agriculture and forestry
Stay Connected
We appreciate your feedback.
Join our newsletter
Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or Research Executive Agency. Neither the European Union nor the granting authority can be held responsible for them.