Weed Identification in Plant Rows Using Feature Extraction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current plant classification algorithms, particularly those using deep learning, are computationally intensive and require large image databases, making them inefficient for real-time weed identification in agricultural settings.

Innovation Solution

A method utilizing a computing unit to receive image information from an agricultural area, identify plant rows, define crop and weed areas, and determine characteristic feature values for plant features, allowing for accurate weed identification within a defined row without the need for extensive image databases.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If deep learning algorithms are used for plant classification, then classification accuracy is improved, but computational complexity and resource requirements increase significantly

Engineering Contradiction:
Improveclassification accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the image processing task into distinct stages: image acquisition, preprocessing (normalization, filtering), feature extraction (color, texture, shape features), and classification. This segmentation allows each stage to be optimized independently, reducing overall computational complexity while maintaining accuracy

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts specific plant features (color histograms, texture features, shape parameters) from images and uses these extracted features for classification rather than processing entire high-resolution images. This extraction approach significantly reduces computational requirements while preserving classification accuracy

Inventive Principle:
Principle #2Taking out (Extraction)

2Measurement precision

If large image databases are used for training, then classification accuracy is improved, but data storage and processing requirements increase

Engineering Contradiction:
Improveclassification accuracyVSAvoidimage data volume
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent uses lightweight, easily obtainable image data from standard agricultural cameras rather than requiring large databases of specialized images. The feature extraction methods work effectively with smaller datasets, reducing storage and data collection requirements

Inventive Principle:
Principle #27Cheap short-living objects (Disposable)

Solution Approach 2:

The patent transforms image data into feature space representations (color histograms, texture features, shape parameters), changing the parameters from raw pixel data to meaningful plant characteristics. This transformation reduces data volume while maintaining or improving classification effectiveness

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If high image resolution is used, then plant feature detection accuracy is improved, but computational intensity increases

Engineering Contradiction:
Improvefeature detection accuracyVSAvoidcomputational intensity
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent extracts essential plant features (color, texture, shape) at optimized resolution levels rather than processing entire high-resolution images. This selective extraction maintains detection accuracy for key plant characteristics while significantly reducing computational intensity

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies different processing qualities to different regions and features: higher resolution and detailed analysis for key plant regions of interest, and lower resolution for background or less critical areas. This local quality approach maintains accuracy where needed while reducing overall computational load

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP3782467B1Method for identifying weeds within a defined row of plants of an agricultural area
Publication Date: 2022.07.06 ROBERT BOSCH GMBH
  • EP3782467B1 patent drawingFigure 1~2
  • EP3782467B1 patent drawingFigure 3~4
  • EP3782467B1 patent drawingFigure 5

AI summary

The invention relates to a method for identifying weeds (16') within a defined row of plants (18) in an agricultural area, comprising the steps of: receiving image information from a field section (10) of an agricultural area with plants (12) detected by means of an optical detection unit; identifying at least one row of plants (18) in the image information by means of a processing unit; defining a crop area (22) comprising the at least one identified row of plants (18) using the at least one identified row of plants (18) and a weed area (24) distinct from the crop area (22) in the image information by means of the processing unit;Determining characteristic values ​​(26) for plant characteristics (26) of plants (16) in the weed area (24) in the image information using the processing unit, in order to obtain a characteristic value range (28) of plants (16) in the weed area; Determining characteristic values ​​(26) for plant characteristics (26) of plants (14) in the cultivated plant area (22) in the image information using the processing unit; and identifying the plants (16') in the cultivated plant area (22) with characteristic values ​​that fall within the obtained characteristic value range (28) of plants (16) in the weed area (24), in order to identify weeds (16') within a defined plant row (18).