Research on fine-grained classification algorithm for remote sensing image of forest and grassland

The monitoring of grassland resources is an important part of grassland ecological monitoring and management, aiming to understand the basic resources of grassland and construct a "one-map" of grassland resource management, so as to facilitate the comprehensive evaluation of the current status and characteristics of grassland resources and provide strong data support for subsequent work such as scientific environmental protection, restoration, utilization and management. Therefore, grassland resource monitoring must be regarded as a fundamental task. This paper focuses on the problem of forest-grass classification in remote sensing images, and proposes a deep learning algorithm be called CBAM + DeepLab V3+ for accurate classification of vegetation, combined with the method of fine-grained image classification, which improves its semantic segmentation effect in remote sensing image, increasing the mIOU from 62.80% to 63.25%, and is able to distinguish forest and grassland more accurately than traditional DeepLab V3+.