Blue-ZE: Artificial vision based smart solution for weed detection in Blueberries plantations
Spain, Basque Country
In the context of fruit growing and harvesting, a clear demand for farming 4.0 solutions has emerged. In this case, the blueberry growing and harvesting process is mainly manual, to prevent damage and largely depends on its point of maturity and the status of the plant. Therefore, a solution that allows the farmer to evaluate the condition of the plant at different points in his fields would allow him to make the appropriate decision when planning the growing and harvesting precision tasks and also adopt the best preventive method in order to eliminate the weeds.
Blue-ZE experiment’s main objective is to use Artificial Vision and Deep Learning technologies to digitize the task of assessing the condition of Blueberries plantations. The proposed solution, first of its kind, enables the extraction of key knowledge, which is based so far on the experience of farmers, without generational change, and supports growing and harvest planning automation in a near future. Moreover, AGRIA ZE‘s strategic objective is to include this new smart solution as part of its catalogue of kits and tools to be integrated in its electrical vehicles and machinery. Blue-ZE solution will also be compatible with AGRIA ZE’s new Myzetrack data platform, designed to enable autonomous vehicle management and precision agriculture tasks, thus improving their efficiency and sustainability
The digitization of agricultural processes within agriculture sector is still in its infancy. The sector conditions for the application of cutting-edge technologies, such as Artificial Intelligence, IoT, BigData, etc. are complex due to changing environmental conditions, low connectivity and distance from inhabited places. In order to enhance the agricultural sector, Blue-ZE experiment proposes to integrate Computer Vision Analysis on images captured at different control points in the fields, Artificial Intelligence as Deep Learning image segmentation method and Data Management in the solution for blueberries quality assessment and these are some innovative technologies that have never been implemented before, constituting a challenge for the company.
The application of Artificial Vision technologies for the optimization of production in uncontrolled environments (outdoors) for the detection, classification and determination of the state of the plants is a necessary first step in any automation process
Weeds are plants that grow spontaneously in undesirable places and grow faster than cultivated plants, so it is advisable to monitor their appearance and growth to prevent them from developing excessively. To do this, weeding (elimination of weeds) must be carried out.
Weeds can be eliminated in different ways: either by hand with the help of a hoe, by using chemical products such as herbicides, by using mechanical implements, by applying laser directly in the weed, etc…. Hand weeding is the most sustainable but at the same time the most laborious method as it requires much more time and dedication. When there is a large amount of weeds in the soil, it is advisable to use herbicides. Modern methods like laser weeding have proven to be as sustainable as the manual ones, providing all the advantages of the automation, taking into account that labor shortage is one of the biggest problems of the sector.
It must be considered that Blueberry’s main quality indicators associated to consumer acceptability are related to the fruit appearance and texture. The main reason why weeds are considered as undesirable plants is their interference in crop development, being able to substantially reduce crop yields and interfere in the proper and healthy growth of the Blueberries. The negative effects caused by weeds can be of two types: competition and allelopathy.
Today, appearance quality assessment of the weeds is determined subjectively using visual observation. Computer vision analysis is a useful tool to evaluate plant optical properties. An accurate and reliable image-based quantification system for weeds additionally is useful for the automation of the complete harvesting process management. It also serves as the basis for controlling robotic harvesting systems. Quantification of weeds from images is a challenging task due to occlusions, differences in size, illumination conditions and the irregular number of weeds that can be present in an image.
Thus, the main technical challenges that the experiment has to face are:
Challenge 1: Obtaining digital images at the defined control points showing the different status of the weeds.
Challenge 2: Quality estimation directly on site and possible back-office classification, which will allow the segmentation of the weeds depending on the images taken.
Challenge 3: Determination of the optimal time for implementing a solution in order to eliminate the weeds from the Blueberries plants (with pesticides, laser or other methods)
Challenge 4: Design and validation of a smart solution deploying an artificial vision solution for a camera on-board of electric agricultural vehicles to enable blueberry harvesting planning support decision system and process automation in a near future.
Implementation of the Solution
At the beginning of the experiment, and in order to be able to implement the solution, AgriaZE has to analyze the use case scenario and has to take into account all the functional and technical requirements specification for the correct development and performance of Blue-ZE smart solution.
On the one hand, Blue-ZE’s architecture is defined to ensure the detection, classification and determination of the state of the plants, as well as data preparation process methodology, which enables an optimal training of proposed Deep Learning algorithms with the correct generated images dataset and ensure correct treatment and management of big data in agriculture.
On the other hand, a set of components is designed for the visualization and monitoring applications, the management of the smart solution and its integration with electric vehicles systems and AGRIA ZE Myzetrack data platform.