Multi-Scale CNN Training for Plant Cut Region Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial neural networks (ANNs) fail to reliably recognize structural features on plants for precise cutting during automated plant propagation due to the loss of fine structures during image scaling and the complexity of plant-specific cut region identification.
Innovation Solution
A method using a regional convolutional neural network (R-CNN) with convolutional and pooling layers that maintains structural feature information during image scaling, combined with parallel processing of multiple camera perspectives and semi-automatic labeling, and employs self-supervised learning to enhance training efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If images are scaled down during neural network processing to reduce computational complexity, then processing speed improves, but fine structural details are lost
Solution Approach 1:
The patent applies segmentation by dividing the image processing into multiple scales. Instead of processing a single scaled-down image, the system processes the original high-resolution image and multiple downsampled versions simultaneously, segmenting the information extraction task across different resolution levels to preserve fine details while enabling efficient processing.
Solution Approach 2:
The patent introduces an additional dimension by processing images at multiple scales rather than a single scale. This multi-scale approach adds a dimensional aspect to the processing pipeline, allowing the neural network to capture both fine structural details from high-resolution images and broader patterns from downsampled images.
2Adaptability or versatility
If traditional neural networks are used for plant cut region recognition, then general object recognition is achieved, but plant-specific structural features cannot be reliably identified
Solution Approach 1:
The patent applies local quality by training the neural network to recognize that different regions of plant images have different importance and characteristics. The system learns to identify specific local features such as nodes, internodes, and leaf attachment points that are critical for determining optimal cut regions, rather than treating all regions uniformly.
Solution Approach 2:
The patent employs parameter changes by modifying the training process and network configuration to optimize for plant-specific features. This includes using plant-specific training datasets, adjusting loss functions to emphasize biologically relevant features, and tuning network parameters to capture the unique structural characteristics of plant tissues and growth patterns.
3Measurement precision
If manual labeling of training data is performed to improve recognition accuracy, then training precision improves, but time consumption and cost increase
Solution Approach 1:
The patent applies self-service by implementing a semi-automatic labeling system where the neural network assists in its own training data preparation. The system generates preliminary annotations or suggestions that can be automatically accepted or reviewed, reducing the manual effort required for labeling while maintaining high accuracy standards.
Solution Approach 2:
The patent uses copying by leveraging synthetic data generation or copying patterns from existing labeled datasets to create additional training examples. This approach reduces the need for extensive manual labeling by generating realistic training data through computational methods while preserving the essential structural features needed for accurate recognition.
Data Source
AI summary
A method for the computer-aided training of an artificial neural network (ANN) for recognizing structural features on objects, by means of which method identified structural features on objects are recognizable rapidly and reliable. That is achieved by virtue of the fact that a convolutional neural network (CNN) having a multiplicity of neurons is used for the training of an ANN for feature recognition on objects. Said network comprises a multiplicity of convolutional and/or pooling layers for the extraction of information from images of individual objects. In this case, the images of the objects are respectively scaled or scaled up and/or down from layer to layer. During the scaling of the images information about the structural features of the objects is maintained, specifically independently of the scaling of the images.