Multilayer Gaussian Feature Extraction Algorithm for Sky Image Classification

The classification of sky images holds significant importance in the field of meteorology as it enables the accurate prediction of weather patterns and climatic conditions. The multi-view features of super pixels have been employed in many algorithms for detection and recognition of various cloud types. However, these algorithms neglect the local structure of region and the mean information of pixel features. These aspects are crucial in capturing the discriminative information necessary for complex data distributions in the sky images. This paper proposes the Multilayer Gaussian (MLG) algorithm to extract the features of sky images using the Gaussian smoothness standard deviation techniques. This method involves capturing the average information within each local patch by calculating the mean of the pixel values. Hence, the informative feature that facilitates the distinction of cloud formations in sky image primarily resides in the average information included within each localized region. Moreover, it provides a comprehensive representation of the local patch and exhibits resistance to noise. The Kiel and SWIMCAT datasets were used to evaluate the proposed MLG algorithm based on the metrics number of extracted features, similarity values, precision, recall, and f-measure using the decision tree and kernel support vector machine. The experimental results show that the performance of MLG surpassed all benchmark feature extraction algorithms.