| 摘 要: 针对无人机的农林航拍图像配准算法存在特征点识别较少、特征点匹配不精准等问题,提出了一种改 进的离散关键点(Discrete Key point,DISK)算法。该算法首先采用梯度策略DISK算法对特征点进行有效描述;其 次采用基于深度学习局部特征的匹配方法进行预匹配;最后利用RANSAC算法去除离群点,筛选匹配结果。为验 证算法的有效性,选取了几组农业航拍图像的数据集进行实验比较。实验结果显示,与经典的SIFT、Dark feat算法 及原始的DISK算法相比,改进的DISK算法显著提高了匹配精度,匹配精度由41.7%提升至98.9%,充分满足农 林航拍图的匹配需求 |
| 关键词: 农林航拍图;梯度策率;局部特征;图像配准;DISK算法 |
|
中图分类号: TP306
文献标识码: A
|
| 基金项目: 国家自然科学基金(32460440);甘肃省高校教师创新基金项目(2023A-051);甘肃农业大学青年导师基金资助项目(GAU-QDFC-2020-08);甘肃省科技计划资助(20JR5RA032) |
|
| Research on Image Registration of Agricultural and Forestry Aerial Photos Based on Improved DISK Algorithm |
|
DONG Yibo, LIU Liqun
|
(College of Inf ormation Science and Technology, Gansu Agricultural University, Lanzhou 730071, China)
bongyb@st.gsau.edu.cn; llqhjy@126.com
|
| Abstract: Aiming at the issues of limited feature point recognition and inaccurate feature point matching in drone-based agricultural and forestry aerial image registration algorithms, this paper proposes an improved Discrete Keypoint (DISK) algorithm. This algorithm first employs a gradient strategy DISK algorithm to effectively describe feature points; secondly, it uses a deep learning-based local feature matching method for preliminary matching; finally, the RANSAC (Random Sample Consensus) algorithm is utilized to eliminate outliers and filter the matching results. To validate the effectiveness of the algorithm, several datasets of agricultural aerial images have been selected for experimental comparison. The experimental results show that, compared to the classic SIFT and Dark feat algorithms, as well as the original DISK algorithm, the improved DISK algorithm significantly enhances matching accuracy from 41.7% to 98.9%, which fully meets the matching requirements for agricultural and forestry aerial photos. |
| Keywords: agricultural and forestry aerial photos; gradient strategy; local features; image registration; DISK algorithm |
