Dual-Task CNN Plant Species Identification With Pixel-Level Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image analysis methods for weed identification in agricultural fields lack sufficient accuracy and granularity in distinguishing between crop and weed species, leading to inefficient herbicide use and environmental impact.
Innovation Solution
A dual task convolutional neural network (CNN) is employed, combining classification and segmentation tasks with separate loss functions to enhance pixel accuracy in plant species identification, using a combination of manually and automatically annotated training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If standardized CNN topologies are used for semantic segmentation of plants, then the device complexity is reduced and ease of operation is improved, but the measurement precision and manufacturing precision of plant species identification deteriorate due to domain differences and intrinsic complexity of plant segmentation
Solution Approach 1:
The patent applies local quality by making the CNN architecture adaptive to local characteristics of plant images. The system uses domain-specific preprocessing, localized feature extraction, and region-aware segmentation that tailors the network's behavior to the specific visual characteristics of plants at different growth stages and under various lighting conditions, thereby improving identification accuracy without requiring complete redesign of the entire system.
Solution Approach 2:
The patent employs parameter changes by dynamically adjusting CNN hyperparameters such as learning rate, batch size, and network depth based on the specific plant species and imaging conditions. The system also modifies architectural parameters like filter sizes and layer configurations to optimize for plant-specific features, achieving higher precision while maintaining reasonable system complexity.
2Measurement precision
If manual segmentation of plants on images is performed to achieve accurate weed discrimination, then the measurement precision of weed identification is improved, but the loss of time and productivity deteriorate due to the great amount of time required
Solution Approach 1:
The patent applies preliminary action by performing extensive preprocessing of plant images before they are fed into the CNN system. This includes domain-specific augmentations, color space transformations, and feature enhancement techniques that are performed in advance to prepare the data for rapid automated processing. This preliminary preparation enables the automated system to achieve manual-level accuracy without the time cost of manual segmentation.
Solution Approach 2:
The patent replaces the mechanical manual segmentation process with an automated CNN-based system that uses learned features and patterns to identify and segment plants. The system substitutes human visual inspection and manual tracing with algorithmic image processing that operates orders of magnitude faster while maintaining or improving accuracy through domain-specific training and optimization.
3Ease of operation
If the same amount of herbicide is applied per surface area, then the ease of operation is improved and device complexity is reduced, but the loss of substance and environmental harmful factors increase due to unnecessary chemical deposition
Solution Approach 1:
The patent applies segmentation by dividing the field into distinct regions based on plant species identification, separating crop areas from weed areas. This spatial segmentation enables selective herbicide application only in weed-infested zones, reducing overall chemical usage. The system creates segmented maps that guide precision spraying equipment to treat only the necessary areas.
Solution Approach 2:
The patent implements self-service by enabling the agricultural system to automatically identify, classify, and generate herbicide application recommendations without human intervention. The CNN system autonomously processes images, identifies weed species and locations, and produces treatment maps that can directly control spraying equipment, eliminating the need for manual field assessment and decision-making.
4Productivity
If different types and rates of herbicides are used to optimize weed control effectiveness, then the productivity and crop growth are improved, but the device complexity and measurement precision requirements increase for accurate weed quantification and staging
Solution Approach 1:
The patent applies dynamics by creating an adaptive system that adjusts its analysis and recommendations based on real-time field conditions, weed species identified, and growth stages detected. The system dynamically modifies its processing pipeline and herbicide recommendations based on the specific situation, enabling optimized weed control that responds to changing field conditions rather than following fixed protocols.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer-implemented method, computer program product and computer system (100) for identifying weeds in a crop field using a dual task convolutional neural network (120) having a topology with an intermediate module (121) to execute a classification task being associated with a first loss function (LF1), and with a semantic segmentation module (122) to execute a segmentation task with a second different loss function (LF2). The intermediate module and the segmentation module are being trained together, taking into account the first and second loss functions (LF1, LF2). The system executes a method including receiving a test input (91) comprising an image showing crop plants of a crop species in an agricultural field and showing weed plants of one or more weed species among said crop plants; predicting the presence of one or more weed species (11, 12, 13) which are present in the respective tile; outputting a corresponding intermediate feature map to the segmentation module as output of the classification task; generating a mask for each weed species class as segmentation output of the second task by extracting multiscale features and context information from the intermediate feature map and concatenating the extracted information to perform semantic segmentation; and generating a final image (92) indicating for each pixel if it belongs to a particular weed species, and if so, to which weed species it belongs.