Quality Assessment of Logs
This case is about the first delivery of Pinta, our multi-purpose image analyzer, to a demanding application of quality assessment of spruce logs. Pinta was integrated to a larger automation system that also measures the logs using advanced laser technology.
Customer: Teknosavo
End customer: Global paper manufacturer
Products used: Pinta, R&D services
Completed: Spring 2008
Background
Spruce logs and stumps are used as raw material for the thermomechanical pulp manufacturing process in the paper industry.
Spruce is, however, prone to rot and this, together with other defects, impairs the quality of the biomass and paper processed from the wood.
To keep the amount of low quality wood at acceptable levels, quality controls are needed. Traditionally quality assessment is done by first selecting random samples from the incoming wood shipments and then inspecting them manually on outdoor warehouse areas. This process is laborous and sometimes severely complicated by difficult weather conditions.
Problem
The end customer was looking for a way to automate the quality assessment of incoming spruce logs and stumps. We were selected by Teknosavo Oy, the system integrator, to provide the needed image analysis software for the task.
The task was to classify logs as acceptable or rotten based on the appearance of log heads.
Special challenges included wildly varying appearance (dirt, snow, ice, variation in color) of the log heads, depending on the weather conditions, harvest season, time spent in the storage etc.
What is more, the imaging system was far from ideal due to fact that it had to be fitted to the existing cutting station infrastucture, resulting in unevenly lit logs.
Solution
Based on our multi-purpose image classifier, Pinta, we created a two-fold solution. The first part is the log head extractor that segments raw images and accurately cuts log heads from them using color features.
The second part is a classifier that takes in the log head images and labels them as rotten, acceptable, or uncertain based on an advanced multi-feature classifier.
Both parts use the neural networks for classification, intuitive visual training tool for configuration, and an API for communicating with the application software, all standard Pinta features.
We also helped the system integrator in the design and configuration of the imaging system to ensure best possible quality of the raw images.
Result
Despite of some unseen challenges, and thanks to exceptional commitment, expertise, and hard work from all parties, the problem was solved and the quality of the incoming spruce logs is now assessed accurately and automatically as a part of the standard cutting process, instead of the laborous manual inspection.
Some of the key elements to the succes on our side, apart from Pinta, was our experience from demanding real-world applications, and unique expertise in robust multi-feature classification techniques that can tolerate varying quality of input image data.
By the way, rumours have it that the end customer is happy!