Periodic Reporting for period 2 - WeLASER (SUSTAINABLE WEED MANAGEMENT IN AGRICULTURE WITH LASER-BASED AUTONOMOUS TOOLS)
Período documentado: 2022-04-01 hasta 2023-12-31
The demand for increased food production results in a rise in herbicide usage. This is a threat to the environment and can cause harm to non-target plants and insects that are important for the soil. The use of herbicides also has severe social implications because they are harmful to both animals and humans and can lead to severe diseases.
Concerned about these effects, the EU has promoted a decrease in herbicide use by seeking alternative solutions. Therefore, the WeLASER project focused on developing a weed management system based on applying lethal doses of energy to the weed meristems using 1) a high-power laser source, 2) an AI vision system, 3) a laser tool and 4) an autonomous vehicle based on smart navigation methods, IoT, and cloud-computing techniques.
This solution was supported by a plan to involve stakeholders in the project decision-making process (Multi-actor approach), communicate, disseminate, and exploit project results to ensure the project's impact on society, industry, and academia, and address ethical issues.
The action resulted in several subsystems that achieved significant performance indicators independently and showed potential for individual exploitation. However, the integrated weeding system was challenging to characterise because more variables than initially planned had to be considered, and different technical and agricultural issues that arose at the end of the project made field tests hard to conduct: low reliability of some subsystems and the seasonal nature of agriculture.
The project results showed scientific and technical significance and can be considered a step forward in weeding with laser. Furthermore, WeLASER promoted laser weeding as a viable alternative to herbicides through various communication and dissemination activities.
Concerned about these effects, the EU has promoted a decrease in herbicide use by seeking alternative solutions. Therefore, the WeLASER project focused on developing a weed management system based on applying lethal doses of energy to the weed meristems using 1) a high-power laser source, 2) an AI vision system, 3) a laser tool and 4) an autonomous vehicle based on smart navigation methods, IoT, and cloud-computing techniques.
This solution was supported by a plan to involve stakeholders in the project decision-making process (Multi-actor approach), communicate, disseminate, and exploit project results to ensure the project's impact on society, industry, and academia, and address ethical issues.
The action resulted in several subsystems that achieved significant performance indicators independently and showed potential for individual exploitation. However, the integrated weeding system was challenging to characterise because more variables than initially planned had to be considered, and different technical and agricultural issues that arose at the end of the project made field tests hard to conduct: low reliability of some subsystems and the seasonal nature of agriculture.
The project results showed scientific and technical significance and can be considered a step forward in weeding with laser. Furthermore, WeLASER promoted laser weeding as a viable alternative to herbicides through various communication and dissemination activities.
The Multi-actor Approach focused on conducting five stakeholder events, four focus group interviews, and computer-assisted telephone interviews in three countries. These events allowed the consortium to involve stakeholders in defining system characteristics, analysing safety, legal, and economic aspects, and addressing upcoming EC Regulation implications. The participants concluded that WeLASER offers features in line with consumer expectations, but the future implementation needs to be cost-effective, have ample capacity, and be multifunctional.
The project developed a 2 µm-wavelength laser source that produce up to 507 W of continuous output power, which was one of the project's objectives. However, some technical issues related to the system's heat dissipation arose during constant use in the field. After improving these issues, the laser source can be marketed for agriculture or other industrial applications.
The weeding tool relied on two customised 0.5-m wide laser scanners to work on two crop rows, but to facilitate further commercialisation, it was dimensioned to tackle four crop rows (2 m) simultaneously. The final characteristics of the tool were those expected, but some limitations of the targeting appeared due to inaccuracies in the cameras and the scanners. Solving these technical problems will enable the commercialisation of the weeding implement as a standalone weeding tool.
A perception system was built using conventional hardware, and AI-based algorithms were developed for 1) crop and weed identification, 2) weed-meristem identification, and 3) meristem tracking. The hardware and algorithms were tested under laboratory conditions, achieving the expected indicators when the system is well-trained for the plants appearing in a specific field. However, it is well-known that AI-based plant identification is less effective for untrained agricultural fields. Therefore, a commercial device should have a strategy to retrain the AI algorithms for site-specific usage during the first year.
The autonomous robot was based on a commercial mobile platform improved with a new power system to provide the energy requested by the laser sources and chillers.
The smart navigation manager (mission planner and supervisor, robot guiding system, safety system, and human-machine interface) met the expected performance indicators. This system can easily be adapted to other commercial platforms, making it suitable for commercial exploitation.
Cloud computing is essential for the WELASER user interface and is the way to store the information acquired during the missions. The features tested in the field demonstrated that the system's performance was aligned with its intended functioning. This subsystem is ready to be commercialised with the smart navigation manager.
The IoT network provided information from the robot and the environment and achieved the expected features. A derived product is the E-fence, a camera system for surveying autonomous operations for human-robot collaboration and alerting about animal/human intrusions in fields. The E-fence is under a process for intellectual protection.
Most performance indicators regarding communication (newsletter, practice abstracts, website, social media, etc.) were achieved. Dissemination had lower indicators than expected (37% of journal publications and 73% of conferences). However, journal articles received more than twice the expected number of citations. In exploitation activities, the consortium managed to file two patent applications. The exploitation plan was completed for presentation to manufacturers and investors, to whom the systems and subsystems were presented in field days held in three countries.
The project developed a 2 µm-wavelength laser source that produce up to 507 W of continuous output power, which was one of the project's objectives. However, some technical issues related to the system's heat dissipation arose during constant use in the field. After improving these issues, the laser source can be marketed for agriculture or other industrial applications.
The weeding tool relied on two customised 0.5-m wide laser scanners to work on two crop rows, but to facilitate further commercialisation, it was dimensioned to tackle four crop rows (2 m) simultaneously. The final characteristics of the tool were those expected, but some limitations of the targeting appeared due to inaccuracies in the cameras and the scanners. Solving these technical problems will enable the commercialisation of the weeding implement as a standalone weeding tool.
A perception system was built using conventional hardware, and AI-based algorithms were developed for 1) crop and weed identification, 2) weed-meristem identification, and 3) meristem tracking. The hardware and algorithms were tested under laboratory conditions, achieving the expected indicators when the system is well-trained for the plants appearing in a specific field. However, it is well-known that AI-based plant identification is less effective for untrained agricultural fields. Therefore, a commercial device should have a strategy to retrain the AI algorithms for site-specific usage during the first year.
The autonomous robot was based on a commercial mobile platform improved with a new power system to provide the energy requested by the laser sources and chillers.
The smart navigation manager (mission planner and supervisor, robot guiding system, safety system, and human-machine interface) met the expected performance indicators. This system can easily be adapted to other commercial platforms, making it suitable for commercial exploitation.
Cloud computing is essential for the WELASER user interface and is the way to store the information acquired during the missions. The features tested in the field demonstrated that the system's performance was aligned with its intended functioning. This subsystem is ready to be commercialised with the smart navigation manager.
The IoT network provided information from the robot and the environment and achieved the expected features. A derived product is the E-fence, a camera system for surveying autonomous operations for human-robot collaboration and alerting about animal/human intrusions in fields. The E-fence is under a process for intellectual protection.
Most performance indicators regarding communication (newsletter, practice abstracts, website, social media, etc.) were achieved. Dissemination had lower indicators than expected (37% of journal publications and 73% of conferences). However, journal articles received more than twice the expected number of citations. In exploitation activities, the consortium managed to file two patent applications. The exploitation plan was completed for presentation to manufacturers and investors, to whom the systems and subsystems were presented in field days held in three countries.
-The High-power Laser Source is a 2 µm single-mode fibre laser device that can generate a maximum continuous output power of 507 W. It can eliminate irradiated weeds during their early stages of growth (cotyledon stages, two-permanent-leaf stage). This laser source would be unique in the market.
-The Perception System relies on a neural network trained to recognise and differentiate between plants with a mean average precision (mAP) of over 90%. The system can identify the meristem position of detected weeds with a mAP accuracy of over 75%. The target coordinates can be used for other weed control methods (mechanical, spot spraying, etc.). It is the first perception system for laser weeding, differentiating monocot and dicot weeds.
-The laser beam is directed at the meristems using mirror-based laser scanners, and the working areas can be changed via software to fit different row spacings. The safety system of the weeding tool turns off the laser when the housing door is opened and guarantees that the beam remains confined, protecting living organisms.
-The mobile platform is endowed with a power system to supply a laser weeding tool currently unavailable on the market.
-The Smart Navigation Manager can provide autonomous guidance on row crops in GNSS-denied areas. In addition, it provides a map builder, supports communication with the cloud and implements a friendly human-machine interface through the Internet.
-The Cloud Computing System provides a mechanism for collecting and storing data and images for posterior analysis and provides a dashboard for IoT sensor data. These features are uncommon in the current agricultural systems.
-The IoT System produced an innovative modular system for monitoring open-field vegetated areas capable of working as an e-fence that, together with a safety/oriented survey of worksites, also allows the monitoring of crops. Such a system is unavailable in the market.
-The Perception System relies on a neural network trained to recognise and differentiate between plants with a mean average precision (mAP) of over 90%. The system can identify the meristem position of detected weeds with a mAP accuracy of over 75%. The target coordinates can be used for other weed control methods (mechanical, spot spraying, etc.). It is the first perception system for laser weeding, differentiating monocot and dicot weeds.
-The laser beam is directed at the meristems using mirror-based laser scanners, and the working areas can be changed via software to fit different row spacings. The safety system of the weeding tool turns off the laser when the housing door is opened and guarantees that the beam remains confined, protecting living organisms.
-The mobile platform is endowed with a power system to supply a laser weeding tool currently unavailable on the market.
-The Smart Navigation Manager can provide autonomous guidance on row crops in GNSS-denied areas. In addition, it provides a map builder, supports communication with the cloud and implements a friendly human-machine interface through the Internet.
-The Cloud Computing System provides a mechanism for collecting and storing data and images for posterior analysis and provides a dashboard for IoT sensor data. These features are uncommon in the current agricultural systems.
-The IoT System produced an innovative modular system for monitoring open-field vegetated areas capable of working as an e-fence that, together with a safety/oriented survey of worksites, also allows the monitoring of crops. Such a system is unavailable in the market.