To enhance the accuracy and efficiency of remote sensing image classification, a parallel classification methodology based on fuzzy support vector machines (FSVM) is proposed. Fully dilated convolutional networks are employed for remote sensing image feature extraction, while both spatial and channel attention mechanisms are incorporated to capture subtle discriminative features. Then, the extracted features are subjected to dimensionality reduction and fusion processing to improve the efficiency of subsequent processing. Finally, fuzzy theory is introduced to construct a fuzzy support vector machine model. The processed features are fed into the system, where parallel computing is integrated into the fuzzy support vector machine (FSVM) framework for remote sensing image classification, thereby enabling parallelised image classification. The results show that the proposed method has a maximum classification accuracy of 97%, an average accuracy of over 90%, a Kappa coefficient of 0.96, and a maximum classification time of only 4.75s. The space complexity remains around 10GB, indicating strong classification performance.
Youlin Cai (Thu,) studied this question.