Study on Pixel Entanglement Theory for Imagery Classification

Classification of remotely sensed images (RSIs) is prerequisite for further applications. Of the existing RSI classification algorithms, the self-organizing mapping (SOM) neural network is considered as one of the most effective unsupervised classification algorithms. However, the traditional SOM only considers the Euclidean distance between neurons, which cannot reflect the correlation characteristics of neurons. For this reason, a radical algorithm, called pixel entanglement (PE) through self-organizing pixel entanglement neural network (SOPENN) is proposed in this article. First, the pixels in an RSI are considered as called quantum pixels, and all of the pixels in a RSI are considered as called quantum pixel array; then PE coefficient is proposed to mine the quantum entanglement relationship of quantum pixels on the array space; finally, a self-organizing neural network (SONN) is established to stimulate the PE behavior between quantum pixels. The theory mentioned above is applied in RSI classification to verify the performance of the SOPENN through four study areas. The experimental results demonstrate that: 1) the classification accuracy of SOPENN model proposed in this article averagely increases 8.42% when compared with the traditional unsupervised classification methods, Iterative Selforganizing Data Analysis (ISODATA), K-means, and SOM; and Kappa coefficient (KC) averagely increases 0.14; and 2) the classification accuracy and KC from the proposed SOPENN model reached the same level when compared with the supervised classification (support vector machine (SVM) model) in four study areas.

Automated summary

This article proposes a new algorithm for the classification of remotely sensed images (RSIs), called the self-organizing pixel entanglement neural network (SOPENN). The experimental results demonstrate that the SOPENN model outperforms traditional unsupervised classification methods and achieves comparable accuracy to supervised classification models.