Counting the number of trees and obtaining information on tree crowns have always played important roles in the efficient and high-precision monitoring of forest resources. However, determining how to obtain the above information at a low cost and with high accuracy has always been a topic of great concern. Using deep learning methods to segment individual tree crowns in mixed broadleaf forests is a cost-effective approach to forest resource assessment. Existing crown segmentation algorithms primarily focus on discrete trees, with limited research on mixed broadleaf forests. The lack of datasets has resulted in poor segmentation performance, and occlusions in broadleaf forest images hinder accurate segmentation. To address these challenges, this study proposes a supervised segmentation method, SegcaNet, which can efficiently extract tree crowns from UAV images under natural light conditions. A dataset for dense mixed broadleaf forest crown segmentation is produced, containing 18,000 single-tree crown images and 1200 mixed broadleaf forest images. SegcaNet achieves superior segmentation results by incorporating a convolutional attention mechanism and a memory module. The experimental results indicate that SegcaNet’s $mIoU$ values surpass those of traditional algorithms. Compared with FCN, Deeplabv3, and MemoryNetV2, SegcaNet’s $mIoU$ is increased by 4.8%, 4.33%, and 2.13%, respectively. Additionally, it reduces instances of incorrect segmentation and over-segmentation.

In object detection, boundary blurring caused by occlusion and background interference often hinders effective feature extraction. To address this challenge, we propose Edge Aware-YOLO, a novel framework designed to enhance edge awareness and efficient feature fusion. Our method integrates three key contributions. First, the Variable Sobel Compact Inverted Block (VSCIB) employs convolution kernels with adjustable orientation and size, enabling robust multi-scale edge adaptation. Second, the Spatial Pyramid Shared Convolution (SPSC) replaces standard pooling with shared dilated convolutions, minimizing detail loss during feature reconstruction. Finally, the Efficient Downsampling Convolution (EDC) utilizes a dual-branch architecture to balance channel compression with semantic preservation. Extensive evaluations on public datasets demonstrate that Edge Aware-YOLO significantly outperforms state-of-the-art models. On MS COCO, it achieves 56.3% mAP₅₀ and 40.5% mAP_50–95 (gains of 1.5% and 1.0%) with only 2.4M parameters and 5.8 GFLOPs, surpassing advanced models like YOLOv11.