Texture Classification Using CNN and Dimensionality Reduction Techniques
Praveen Kumar V, Jeebananda Panda · 2022
Texture classification has been a popular topic in recent years for researchers. Texture classification frequently employs Convolution Neural Networks (CNN) as the traditional handcrafted methods were not successful for the immense size of data. Handcrafted methods are tedious and costly to develop and highly dependent on one application. Convolution neural networks (CNN) have recently emerged as the most important feature extractor: CNN-based feature extraction methods actually outperforms other feature extraction techniques. In precedent works, this issue of Texture Classification was resolved by deep-learning techniques but the classification accurateness was 96–97 percent. Deep learning actually learn features in incremental manner. But in CNN we deals with millions of parameters. Training it with large and complex data models can be expensive. It also needs extensive hardware. So Deep Learning has very high Dimensionality. So. in my methodology i have used Dimensionality reduction technique after the Feature extraction. So that further reduce the multidimensional to fewer dimension in Visualization. So, In my work, I have proposed the hybrid CNN technique that increases the accuracy as well as that better visualization of the network.