#	Pagina
attuale pagina	/open-h2020/projects/212439/results.html

Report

Teaser, summary, work performed and final results

Periodic Reporting for period 1 - GREENPATROL (Galileo Enhanced Solution for Pest Detection and Control in Greenhouse Fields with Autonomous Service Robots)

Teaser

Summary

GreenPatrol solution is focused on tackling the EU need of producing more with less, as the European population grows while the area available for agriculture is gradually declining due to increased forestry and urbanisation. Greenhouse crop production is a growing reality throughout the world, which shows a 22% accumulated increase in area since 2011. Protected cultivation (greenhouses) emerges as a way to protect crops from adverse weather conditions, allowing year-round production. However, the intensification of greenhouse crop production creates favourable conditions for many devastating pests and diseases that can cause a decrease of about 15% of losses. This has significantly increased the need for pesticide applications. At the same time, legislative measures and requirements regarding the quality and safety of vegetables have become increasingly demanding. Consumers awareness has raised and thus the demand for pesticide-free products.
GreenPatrol aims at developing an innovative and efficient robotic solution for Integrated Pest Management in crops, which has the ability to navigate inside greenhouses while performing pest detection and control tasks in an autonomous way. The GreenPatrol navigation capability is enabled by Galileoâ€™s new signals and the implementation of sensor fusion techniques.
The main overall objectives include: 1) Robot precise positioning solution able to operate in the intended challenging environment, providing accurate and detailed pest maps for decision making about precise case-specific treatment, 2) Integration of Galileo better capabilities in light indoor environment, 3) Perception with visual sensing for on-line pest detection, including reasoning mechanisms for efficient action selection and 4) Control strategies for manipulation and motion planning based on pest monitoring system feedback.

Work performed

During the first period, system specifications were determined based on interviews to potential users, stakeholders and analysis of greenhouses.
The system architecture has been designed from a functional and modular point of view. Tests have been defined in order to verify the fulfilment of each requirement at subsystem level, and for the whole system. Safety mechanisms have also been considered both at hardware and software levels.
Several measurement campaigns in greenhouse have been carried out to select and to assess the performance of sensors: COTS GNSS specifications, hardware and software requirements, IMU and odometer integration, range laser sensors, relative localization strategies, pest detection and identification approach, vision sensors, types of algorithms, image dataset requirements for training and validation, etc. Technology gap has been studied for each of the main functionalities of the system: IPM strategies, pest identification, Localization and autonomous Navigation. It has been concluded the viability of the autonomous navigation in the greenhouse and the clear advantages of the use of Galileo E5 AltBOC signals in terms of signal quality and positioning performance.
GreenPatrol Localization subsystem is composed of an absolute localization module using precise positioning GNSS techniques combined with inertial sensors and odometry (PPP-DR) and a relative localization module that exploits robot range readings and generates a model of the environment to estimate the robot position inside it. Both modules have been verified using data gathered in real greenhouse environment and mock-up in several campaigns.
Autonomous Navigation algorithms have been developed, trained and verified in simulation environment, concluding that the proposed solution provides the required performance.
GreenPatrolâ€™s manipulative capabilities have been designed and implemented endowing with strategies for adaptive path planning and control based on sensor information, offering the needed functionalities to perform the pest detection and treatment operations. Work has been done in the selection of the hardware for pest inspection tasks. A tomato disease dataset for the target pests has been generated using manual and automatic systems and including the different tomato diseases features. Three different approaches (computer vision, machine learning and deep learning) have been tested in order to select the best performing one in terms of detection, identification and early pest detection accuracy.
From tests in cultivation chambers and real environment, a first proposal of the IPM strategies has been defined in order to execute the sequence of pest detection and pesticide application operations. Additionally, a data storage/management system has also been implemented, ready for local and cloud solutions and able to cope with several mobile manipulators.
At the end of the reporting period, all the subsystems have been tested in laboratory and real scenario conditions, and the integration process for sensors and actuators in the robot platform is ongoing.

Final results

The main advance in GreenPatrol project is the introduction of a robotic platform for autonomous and effective scouting and pest control based on an Integrated Pest Management (IPM) decision support system. Unlike most of the robotic solutions for pest detection and treatment (mobile bases including inspection systems), GreenPatrol has a manipulator arm on a mobile base, that provides a higher mobility to the inspection system and is able to analyse plants in different situations and plant growth stages.
Precise smart agriculture entails considerable demands in the development and implementation of new processes, addressed by this project. It is targeted the use of artificial intelligence and the development of an autonomous robot able to early detect and evaluate potential crops pests. However, the development also brings considerable demands on completely new procedures of IPM strategy for robotized agriculture, which is another goal. A huge number of images is needed to generate Artificial Intelligence detectors, that was not available before the project. Sharing the so far generated dataset will be very beneficial for the scientific and agriculture communities.
Currently, there is no automatic method used for directly monitoring plants for pest detection in greenhouse crops using a robotic platform. The proposed automated identification based on computer vision will provide an objective measure to evaluate the pest status and possible extension. All data generated during the pest detection task is stored using a data store/management system that will be very useful to develop prediction models based on environmental conditions.
Most robots devoted to greenhouses are specialized to particular crop and environment. For navigation, it is common to rely on rails or heating pipes or to require heavy environment instrumentation, such as beacons, tags or landmarks. In the case of autonomous robots, the precise topology of the greenhouse (number of rows, length, separation, etc.) is often required. GreenPatrol robot provides a fully autonomous solution able to navigate without the aforementioned restrictions. It uses a combination of global and local information to generate detailed maps of the environment, to locate itself and the detected pests. The global localisation uses Precise Point Positioning techniques to get high accuracy in any location, and the Galileo E5 AltBOC signal to help improve tracking and reduce multipath in the greenhouse environment.