An integrated approach of Active Incremental fine-tuning, SegNet, and CRF for cutting tool wearing areas segmentation with small samples

Cutting tool wear is a critical factor that affects product quality in manufacturing processes and measuring the flank wear area is the most common method to assess the condition of the tool. Nowadays, the direct way based on image processing has been developed due to its high information content and does not rely on expensive subsidiary measurement equipment compared with the indirect way. However, the direct way, whether based on computer graphics or based on artificial intelligence, has a shortcoming. The traditional computer graphics methods have poor robustness, and the artificial intelligence way enabled with deep convolutional neural networks (DCNN) requires a large amount of data and it usually works well on its training data. This article proposes a new architecture based on active incremental fine-tuning, SegNet, and CRF. The new architecture integrates active incremental fine-tuning and conditional random field with an optimized SegNet. The new architecture has greatly improved the running speed and reduced the model size. Moreover, the architecture can be trained with small samples and obtain high precision. Finally, in our case, the architecture achieves an average accuracy rate of about 88% on a small dataset. The training process consumes about 4612 s, and the number of learning parameters is reduced to 788,006. The methodology in the article has been verified through experiments.