Search

RU-2861308-C1 - DEVICE FOR REMOVING WEED PLANTS

RU2861308C1RU 2861308 C1RU2861308 C1RU 2861308C1RU-2861308-C1

Abstract

FIELD: agriculture. SUBSTANCE: device for removing weed plants comprises a frame (7) on which are mounted: an undercarriage (2), an engine, a battery, a generator, a control unit (1) with an onboard computer (12) and a recognition unit in the form of a camera (4), a navigation module (11), a manipulator (6) with a control device (8) and a tool for mechanical removal of weed plants, a container (9) for collecting weed plants, and a collision avoidance block mounted at the lower level of the device. An algorithm of a detection system is loaded into the onboard computer (12), configured to recognise and classify images of weed plants, to compare current images of weed plants with previous ones and to transmit the result to a trained neural network to determine the type of weed plant. The neural network is configured to send a signal to the control unit (1) to start the manipulator (6), comprising pivotally connected links, fingers (3) and a vacuum pump, is configured to bend and rotate in different directions, grip, tear off and place weed plants into the container (9). The fingers (3) of the manipulator (6) are made of polymer and coated with relief rubberised pads. The device also comprises an ultrasonic sensor and a gyro sensor (13) mounted on the frame, as well as an inertial measurement unit mounted at the lower level of the device, with the ability to determine the position of the device in space, its tilt and rotation speed. EFFECT: increasing the quality, increasing the reliability of weed plant removal, reducing labour costs and simplifying the weed removal process, since the device allows eliminating the human factor, automating and digitising the process of weed plant removal. 9 cl, 5 dwg

Inventors

  • SEITOV SANAT KAIRGALIEVICH
  • Shermatova Ajzhamal Kamchybekovna

Dates

Publication Date
20260504
Application Date
20250127

Claims (9)

  1. 1. A device for removing weeds, comprising a frame on which are installed: a chassis, an engine, a battery, a generator, a control unit with an on-board computer and a recognition unit made in the form of a camera, a navigation module, a manipulator with a control device and a tool for mechanically removing weeds, a container for collecting weeds, and a collision prevention unit installed at the lower level of the device, characterized in that an algorithm of a detection system is loaded into the on-board computer, configured to recognize and classify images of weeds, with the ability to compare current images of weeds with previous ones and transmit the result to a trained neural network to determine the type of weed, wherein the neural network is configured to send a signal to the control unit to start the manipulator, wherein the manipulator, containing hinged links, fingers and a vacuum pump, is configured to bend and rotate in different directions, grab, tear off and place weeds in a container, wherein the fingers of the manipulator, made of polymer, are covered with embossed rubberized overlays, and the device also contains an ultrasonic sensor and a gyro sensor mounted on the frame, as well as an inertial measurement unit mounted on the lower level of the device with the ability to determine the position of the device in space, its tilt and rotation speed.
  2. 2. The device according to paragraph 1, characterized in that the chassis is tracked.
  3. 3. The device according to paragraph 1, characterized in that a convolutional neural network is used.
  4. 4. The device according to paragraph 1, characterized in that the mechanism for bending the fingers of the manipulator includes four drives, fixed to the tip of the manipulator, connected by rods to the fingers and driving the fingers into action, wherein the fingers of the manipulator are connected to the tip of the manipulator by means of four hinges, each finger containing three hingedly connected parts.
  5. 5. The device according to paragraph 1, characterized in that the container on the upper part of the front wall contains an infrared sensor that notifies about the container being full of weeds and about the need to unclench the fingers of the manipulator when the weeds are above the container, wherein the unclenching of the fingers is carried out by means of return unclenching springs located inside the fingers.
  6. 6. The device according to claim 1, characterized in that the collision avoidance unit includes sonars.
  7. 7. The device according to paragraph 1, characterized in that the device is equipped with rotation angle sensors.
  8. 8. The device according to claim 1, characterized in that the device additionally contains a single-board computer that receives a video stream and photographs of plants from a camera in real time and passes it through a library of computer vision algorithms.
  9. 9. The device according to paragraph 1, characterized in that the on-board computer is designed with the ability to record the coordinates of the areas of the field in which the manipulator failed to pull out the weed or only partially pulled it out.

Description

The invention relates to agricultural engineering, in particular to a robotic vehicle designed for the automatic removal of weeds. The prior art provides technical solutions for removing or destroying weeds. Existing methods of weed control (hereinafter referred to as weeds) are primarily chemically oriented, while environmentally friendly methods are not available to the general user and are not widely used in Russia. Among gardeners, thorough manual weeding remains the most accessible and reliable method of weed control, with the best results achieved by removing plants at the seedling stage. Besides chemical weed control, mechanized weed control is also very popular. This method involves compacting the soil when heavy equipment (powerful tractors with harrows, ploughshares, and disc cultivators) passes through. Harrowing destroys 1-2% of crops in a field, along with weeds, as this method does not distinguish between different plant species. Automated, highly accurate weed recognition systems based on images, based on modern approaches based on deep machine learning, are becoming increasingly relevant. These research challenges are particularly relevant for organic agriculture, which does not use herbicides and is therefore vulnerable to weeds. Weed identification is complicated by the fact that the visual appearance of different plant species often differs only in minor details. Digital cameras and their accompanying software are commonly used for machine vision, as are traditional solutions based on digital video cameras. These provide a highly reliable, easily configurable, and easy-to-install coordinate determination system. The mission of automated devices should be to eliminate or minimize the use of herbicides in crop cultivation. Eliminating herbicides through mechanical weed removal is aimed at protecting the soil and plant products from contaminants and residues. Currently, most fields on large farms worldwide are sprayed with herbicides. Their effects affect the entire field, including those parts of the plant that are intended for human consumption. Crops are generally not killed by herbicides, but even with the right chemicals, their growth is inhibited, which ultimately negatively impacts the yield. Weeds are developing resistance to herbicides, necessitating the development of new, more effective herbicides. Weeds also develop resistance to these new herbicides, creating a vicious cycle in which agricultural producers race to find new herbicides, the effectiveness of which is blocked by the development of resistance in weeds. Innovators around the world are trying to solve or at least reduce the impact of the above problems. The BoniRob robot [BoniRob is an agricultural robot from Bosch that fights weeds by pushing them back into the ground. URL: https://www.nanonewsnet.ru/articles/2015/bonirob-selskokhozyaistvennyi-robot-kompanii-bosch-kotoryi-boretsya-s-sornyakami-zabiv (accessed on 02.02.2023)] is designed to find and destroy weeds in the field. The Tertill robot [Tertill. URL: https://tertill.com/ (accessed on 02.02.2023)] prevents weeds from growing by cutting them off before they reach 2.5 cm. Y. Xion, Y. Ge, Y. Liang, and S. Blackmore have created a robot that kills weeds with laser light. Plant images are processed in two stages. The first stage involves distinguishing between plants and their surroundings by converting the image into a binary image: plants are assigned white pixels, while the rest of the background is black. Based on the color differences in the photographs, the developers use an algorithm that compares the hue, saturation, and brightness of each pixel in the image with a specified color gradation inherent in the plants and surrounding area. This distinguishes between the plants and their surroundings. Furthermore, the algorithm learns to distinguish colors by taking into account changes in the environment and light levels throughout the day. Next, the images undergo median filtering to remove soil noise and similar unimportant details. Gaps in the plant images caused by highlights are also artificially filled to ensure accurate center-of-mass calculations. The second stage is plant segmentation, which allows weeds to be distinguished from other crops. The authors use an image "blurring and expanding" method to distinguish weeds from crop plants based on differences in leaf shape. The method involves sequentially removing and restoring image features, while comparing the processed images step by step. Ultimately, the images retain specific details unique to weeds: for example, the specific shape of leaves, their arrangement on the stem, and so on. The image processing time using the "blurring and expanding" method is shorter—0.0251 seconds—than other methods, which is why it was chosen. The device destroys detected weeds using laser radiation. The effectiveness rate is 97% (meaning the proportion of weeds destroyed out of the total number in a given field area). Ho