基于改进YOLOv3的单木树冠检测算法  

Individual Tree Crown Detection Algorithm Based on Improved YOLOv3

作　　者：王丽文 朱正礼[1] 云挺[1] WANG Li-wen;ZHU Zheng-li;YUN Ting(College of Information Science and Technology,Nanjing Forestry University,Nanjing Jiangsu 210037,China)

机构地区：[1]南京林业大学信息科学技术学院,江苏南京210037

出　　处：《计算机仿真》2023年第1期510-516,共7页Computer Simulation

摘　　要：针对已有目标检测算法在检测形态结构复杂的树冠时容易产生误检和复杂度较高的问题,提出一种基于改进YOLOv3的单木树冠检测算法。结合稀疏连接与残差结构的特点,采用ResNeXt-50模型构建基础网络,并根据ResNeXt块设计检测模块的结构,降低网络复杂度,改善网络泛化性与表征能力。同时采用Mish激活函数改进隐藏层的映射形式,使网络更易于拟合和优化。实验结果表明,与YOLOv3算法相比,上述算法在航拍树冠数据集上的平均精度均值提高了9.07%,且具有更小的时空复杂度。To address the problem that the existing object detection algorithms are prone to false detection and have high complexity when detecting the tree crown with complex morphological structure, a detection algorithm for individual tree crown based on improved YOLOv3 is proposed. Combined with the characteristics of sparse connection and residual structure, the ResNeXt-50 model was used to construct the basic network, and the structure of the detection module was designed according to the ResNeXt block to reduce the network complexity and improve the generalization and feature representation capability of the network. At the same time, the Mish activation function was used to improve the mapping form of the hidden layers, so that the network is easier to fit and optimize. Experimental results show that, compared with the YOLOv3 algorithm, the proposed algorithm improves the mean Average Precision(mAP) on the aerial tree crown dataset by 9.07%,and it has a smaller space-time complexity.

关 键 词：深度学习 目标检测 单木树冠 航拍图像 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]