A novel type of feature extraction algorithm based on Bandelet transform and local binary pattern for visual optical electronic nose

• Visual optical e-nose response images possess complex feature characteristics. • A novel feature extraction algorithm based on Bandelet-LBP is constructed. • Structural, periodic, and directional features can be effectively acquired. • The algorithm enables 93% odor identification accuracy in visual optical e-nose. To address the characteristics of visual optical electronic nose (E-nose) response images - including a huge amount of data, multiple frequency components, and complex spatial-temporal features, a novel type of feature extraction algorithm based on Bandelet transform and local binary pattern (Bandelet-LBP) for visual optical E-nose was proposed. Firstly, the gas sensing mechanism of the visual optical E-nose was introduced. Next, a detailed description of the theoretical model of the texture image feature extraction algorithm based on Bandelet-LBP. Then extracts the features of the response images of different gases using the local binary pattern (LBP), the gray-level co-occurrence matrix (GLCM), the Gabor wavelet, and the Bandelet-LBP, respectively. Finally, the classification algorithms such as correlation coefficient (CC), euclidean distance to centroids (EDC) and support vector machine (SVM) were chosen to pattern analyze of the gas feature data. Experimental results showed that compared with classical LBP (recognition accuracy 77.65%), GLCM (recognition accuracy 84.52%), and Gabor wavelet (recognition accuracy 88.21%), the Bandelet-LBP achieved a recognition accuracy of 92.95%, which confirmed the effectiveness and superiority of the Bandelet-LBP algorithm.