Lumber Grading via Deep Learning Semantic Segmentation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing wood characteristics detection systems require extensive human expertise and are inefficient in accurately identifying features like knots, which are crucial for lumber grading, due to their complexity and variability in orientation and presentation.

Innovation Solution

The method employs deep learning techniques using the Caffe framework and SegNet architecture for semantic segmentation, enabling automatic feature selection and robust classification of wood characteristics, such as knots, without human intervention, by training convolutional neural networks on large datasets.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional image processing techniques and rules are used to detect wood characteristics, then the system requires extensive human expertise and manual feature extraction, but this leads to difficulty in accurately identifying features like knots due to their complexity and variability

Engineering Contradiction:

Improveaccuracy of knot detectionVSAvoidcomplexity of feature extraction process

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces manual mechanical feature extraction and expert rule-based systems with deep learning neural networks that automatically learn and extract features from images. The convolutional neural network performs automatic feature selection and robust classification of wood characteristics without requiring human expertise in feature engineering, thereby improving detection accuracy while reducing system complexity.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The deep learning system is self-training and self-improving, automatically learning from large datasets of wood images to identify knots and defects. The system performs its own feature extraction and classification tasks without continuous human intervention, adapting to the variability of wood characteristics through automated learning from training data.

Inventive Principle:

Principle #25Self-service

2Productivity

If experts manually extract features using rules and image processing, then some level of detection is achieved, but the process is difficult, time consuming, and not guaranteed for accuracy

Engineering Contradiction:

Improvespeed of lumber gradingVSAvoidtime for feature extraction and selection

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The patent replaces time-consuming manual feature extraction with automated deep learning systems that process images rapidly. The convolutional neural network performs feature extraction, selection, and classification in an integrated automated process, eliminating the sequential manual steps and significantly reducing the time required for lumber grading while maintaining or improving accuracy.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Reliability

If computer vision systems are programmed to identify knots with explicit rules, then specific knot patterns can be detected, but the system becomes fragile and needs constant improvement and maintenance

Engineering Contradiction:

Improverobustness of detection systemVSAvoidability to handle various knot orientations and presentations

Core Design Contradiction:

ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent implements a dynamic, adaptive system using deep learning neural networks that continuously learn from training data. Unlike static rule-based systems, the neural network adapts to various knot orientations, sizes, and presentations by learning from diverse training examples. This dynamic learning capability makes the system robust and versatile without requiring constant manual updates or maintenance.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The deep learning system changes its internal parameters (weights and biases) automatically during training to adapt to different knot patterns and presentations. This parameter optimization through learning enables the system to handle variability in knot orientations and appearances, improving both reliability and adaptability compared to fixed rule-based approaches.

Inventive Principle:

Principle #35Parameter changes

4Measurement precision

If a minimum set of extracted features is selected to maximize accuracy, then detection precision may be improved, but the effort to derive and select features is difficult and time consuming

Engineering Contradiction:

Improvedetection accuracyVSAvoidtime for feature derivation and selection

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent replaces the manual process of feature derivation and selection with automated deep learning systems. The convolutional neural network automatically identifies and extracts the most relevant features from raw images during training, eliminating the time-consuming manual feature engineering process while maintaining or improving detection accuracy through data-driven feature selection.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS10830755B2Board lumber grading using deep learning semantic segmentation techniques

Publication Date: 2020.11.10 MICROTEC INC

AI summary

A method of board lumber grading is performed in an industrial environment on a machine learning framework configured as an interface to a machine learning-based deep convolutional network that is trained end-to-end, pixels-to-pixels on semantic segmentation. The method uses deep learning techniques that are applied to semantic segmentation to delineate board lumber characteristics, including their sizes and boundaries.