Soil Image Navigation Using CNN Descriptors Without GPS Wires
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous lawn mowers and agricultural machines require external infrastructure like wires or GPS systems for navigation, which are complex, costly, and limited in portability, and existing image processing methods are not robust enough to handle environmental perturbations such as shadows, lighting changes, and obstacles.
Innovation Solution
A method using a trained convolutional neural network for image processing that generates a synthetic descriptor of the soil, allowing the machine to autonomously navigate and perform tasks without external infrastructure, by learning soil characteristics and being robust to perturbations through hierarchical image representation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If external infrastructure like GPS systems or wires is used for navigation, then navigation accuracy and control are improved, but device complexity and cost increase
Solution Approach 1:
The machine equips itself with image acquisition means and processing units to perform autonomous navigation and soil analysis without external infrastructure. The system captures images, processes them through convolutional neural networks, and generates control signals independently, making the machine self-sufficient rather than dependent on external GPS or wire systems.
Solution Approach 2:
The patent replaces traditional mechanical or electronic navigation systems (GPS antennas, peripheral cables) with an optical-based image processing system. Instead of using radiofrequency signals or physical guides, the machine uses cameras and neural networks to perceive and navigate the environment, substituting optical information processing for conventional navigation mechanisms.
2Reliability
If external infrastructure like peripheral cables or beacons is deployed, then machine confinement and trajectory control are improved, but ease of operation and portability deteriorate
Solution Approach 1:
The machine independently acquires images of its environment and processes them to determine its position and generate navigation commands without requiring external beacons or cables. The system performs all processing functions onboard, enabling it to operate in diverse locations without infrastructure deployment.
Solution Approach 2:
The patent extracts the navigation and analysis functions from external infrastructure and consolidates them within the machine itself. By removing the dependency on external cables, beacons, or GPS systems, the machine becomes portable and adaptable to different working areas without requiring infrastructure installation or configuration.
3Ease of manufacture
If traditional image processing methods are used, then implementation simplicity is improved, but reliability under environmental perturbations deteriorates
Solution Approach 1:
The system performs preliminary training of convolutional neural networks using large datasets of images captured under various environmental conditions (different lighting, weather, soil types, obstacles). This pre-training enables the network to learn robust feature representations that generalize well to new, unseen conditions, improving reliability without increasing operational complexity.
Solution Approach 2:
The patent employs a composite approach by combining multiple convolutional neural networks with different specialized functions (soil classification, obstacle detection, navigation). Each network is trained for specific tasks, and their outputs are integrated to produce comprehensive control signals, creating a robust multi-functional system that handles various environmental perturbations effectively.
Data Source
AI summary
The invention relates to a method for controlling a soil working means, based on an image processing. Such a soil working means comprises a locomotion member and a working member. The method comprises the steps of acquiring at least one digital image of the soil by means of digital image acquisition means installed on the working means; processing, by means of an electronic processing unit, the at least one digital image acquired by performing at least one convolution operation on the digital image by means of a trained neural network; obtaining, by means of the electronic processing unit, at least one synthetic soil descriptor based on such a processing; generating, by means of the electronic processing unit, at least one control signal of the locomotion member or of the working member based on the synthetic soil descriptor.