Enhancing Wood Quality Assessment with Automated Classification Systems
Introduction
A key challenge faced by companies operating in the wood industry is the accurate classification of wooden sheets based on their visual quality. The traditional approach to wood quality assessment involves visual inspection by human specialists, who rely on their experience and judgment to make subjective classifications. This process, however, has some major drawbacks:
- Time Consumption: A human will always be slower than a machine to perform this task, especially when dealing with borderline cases.
- Availability of Human Labor: When an employee is absent from work, quits, or is being trained on the job, productivity decreases.
- Inconsistencies and Errors: Two different specialists may disagree on the classification of a borderline case, resulting in lost revenues when high-quality wood is misclassified.
Wood quality assessment is being revolutionized by recent advancements in the field of computer vision. A higher degree of automation at the inspection stage reduces risks for wood manufacturers, enabling budget forecasts to be met consistently. The next section highlights the main components of the solution developed by Ennova Research and how such a solution improves the finances of its customers.
Proposed Solution
We proposed a comprehensive Machine Learning pipeline that not only enhances quality assessment but also minimizes human error and subjectivity:
- Data Collection and Expert Annotations: Human specialists label the data as part of their routine work.
- Inter-Annotator Agreement: Only the sheets where experts reach a consensus on wood sheet quality are selected for further analysis, reducing ambiguity and noise in the data.
- Model Training or Fine-Tuning: With the data prepared, our AI model is either trained from scratch or fine-tuned if an existing model is available.
- Deployment of Improved Model: If the newly trained or fine-tuned model outperforms its predecessor, it is deployed into production.
- Iterative Improvement: The previous steps are repeated until the model cannot improve anymore.
- Human Supervision for Borderline Cases: Borderline cases, which the AI finds difficult to classify, are supervised by human employees and their annotations are added to the training and test sets to further refine the model capabilities.
Results
By harnessing the capabilities of Google Cloud Platform AutoML Vision, we have achieved a state-of-the-art wood sheet classification model that has significantly elevated the standards of quality assessment for the customer. The integration of Google Cloud Platform AutoML Vision, in particular, proved game-changing. The model’s efficiency in automating the classification of wooden sheets allowed us to focus on creating the whole pipeline and prepare it for the scale-up phase, in which dozens of types of woods will be managed.
