Comparative Study of CBAM-CNN and Other Deep Learning Algorithms for Remote Sensing Imagery Extraction of Agricultural Greenhouse

Accurate extraction of the spatial distribution information of agricultural greenhouses, a vital component of modern agriculture, is of significant importance for agricultural management and environmental monitoring. This study took Yaojia Town in Zhongmu County, Zhengzhou City as the research area, utilized high-resolution GF-2 satellite imagery acquired on January 4, 2024 as the primary data source, and investigated the capability of different machine learning models for agricultural greenhouse extraction. Four classification models were constructed: convolutional neural network (CNN), random forest (RF), support vector machine (SVM), and an optimized convolutional neural network incorporating an attention mechanism (CBAM-CNN). These models were applied to extract agricultural greenhouse information from the study area. The classification results from each method were qualitatively and quantitatively compared and analyzed using visual interpretation data and confusion matrices. The results indicated that the CBAM-CNN model achieved the best extraction performance, with an overall accuracy of 94.26% and a Kappa coefficient of 91.24%. The resulting classified patches were complete with clear boundaries, demonstrating the strongest robustness in suppressing salt-and-pepper noise and background interference. The CNN and RF models followed, with overall accuracies of 91.46% and 86.24%, respectively, achieving good extraction results but slightly inferior to CBAM-CNN. The SVM model yielded the lowest overall accuracy at 80.25%, with notable misclassifications in its results. This study validates the effectiveness and superiority of deep learning models incorporating attention mechanisms for the high-precision extraction of agricultural greenhouses from high-resolution remote sensing imagery, providing a scientific methodological foundation for intelligent monitoring and management of agricultural resources.