Accurate extraction of forest resources and their distribution is crucial for the management, protection, and sustainable utilization of forests. In order to address the issues of low accuracy and poor edge definition in existing methods of forest extraction, this paper introduces a novel approach by using remote sensing images called the Context Awareness and Boundary Constraints Network (CABC-Net). Initially, a Context Awareness (CA) module is developed to explore the interconnections among spatial pixel data, extracting ample global contextual information through the transfer of features across layers. This reduces the impact of contextual variations and complex backgrounds on the detection outcomes. Secondly, to further enhance the edge precision, a Boundary Constraints (BC) module is proposed that integrates boundary features with deeper network features to stabilize the localization of forest boundaries and refine areas of uncertainty, thus boosting edge discernment. Finally, to verify the validity of the method, a new dataset is created and case test is analyzed. The results indicate that the Intersection over Union (IoU) and pixel accuracy (PA) of this method are improved by 0.55% to 9.45%, and by 0.19% to 7.53%, respectively. The analysis demonstrates that the proposed  method has better boundary integrity in the face of complex scenes, and can be better applied to forest land extraction.