Plant Classification Recheck for Edge-Aware Field Spraying
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Agricultural spraying systems face challenges in accurately classifying plants at the edge of the image due to partial visibility, leading to misclassification and incorrect application of sprays, especially when data flow between cameras is limited and computing power is restricted, as in autonomous field sprayers.
Innovation Solution
The method involves using an optical detection unit inclined relative to the field to capture trapezoidal images, classifying plants into two classes, and reclassifying plants from the 'weed' class to the 'cultivated' class if they extend beyond a defined boundary line from the image edge, ensuring accurate classification and application of sprays.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If plants at the image edge are classified using standard algorithms, then classification is performed, but misclassification occurs leading to incorrect spray application
Solution Approach 1:
The patent applies different classification rules to different spatial zones within the image. Specifically, plants detected in the rear image section (within the reclassification zone) are subject to reclassification logic, while other plants use standard classification. This local differentiation resolves the contradiction by improving classification accuracy for edge plants without affecting the reliability of spray decisions for fully visible plants.
Solution Approach 2:
The system performs preliminary reclassification of plants in the reclassification zone before final spray decisions are made. By pre-identifying and reclassifying potentially misclassified edge plants, the system ensures that subsequent spray applications are based on accurate classifications, thus resolving the contradiction between detection and spray reliability.
2Measurement precision
If reclassification is performed for all plants, then classification accuracy improves, but computing time and power increase excessively
Solution Approach 1:
The patent restricts the reclassification process to only those plants located in the rear image section within the defined reclassification zone. This localized approach means that not all plants undergo the computationally intensive reclassification process, thereby maintaining improved accuracy for edge plants while avoiding excessive computing time and power consumption that would result from processing all plants uniformly.
3Area of stationary object
If the optical detection unit is inclined to capture the field, then the field of view is optimized, but plants at the rear image edge are partially visible and may be misclassified
Solution Approach 1:
The patent acknowledges that the inclined optical detection unit creates a trapezoidal image with varying plant visibility across different zones. By applying reclassification logic specifically to plants in the rear image section where partial visibility causes misclassification, the system compensates for the geometric distortion without requiring a change to the optimal inclined detection configuration.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
The invention relates to a method for classifying plants (22, 23, 60), particularly for agricultural purposes, comprising the following steps: - capturing a field portion (34) having plants (22, 23, 60) by means of an optical capturing unit inclined with respect to the field portion (34), in order to obtain image information (36) having a rear image edge (44) with an image width B(image,h) and having a front image edge (46) with a greater image width B(image,v); - classifying the plants (22, 23, 60) in the obtained image information (36) into at least one first plant class and into at least one second plant class by means of an algorithm in order to treat, in accordance therewith, plants (23, 60) of the second plant class with a medium, more particularly liquid spray agent (20); wherein the step of classifying comprises a step of reclassifying, in which classified plants (221) of the second plant class are reclassified into the first plant class by means of the algorithm if said plants extend, in a rear image portion (52) of the image information (36) which is defined by means of a defined image separation line (50), from a lateral image edge (48) of the image information (36) at most to a defined reclassification boundary line (54).