| 摘 要: 针对同种藻类形态各异、藻类相似性高、采集图像中存在气泡和杂质等带来的检测与识别准确率低的问题,提出了基于多方法融合的淡水水域常见藻类检测与识别算法。首先利用颜色信息识别藻类图像的背景颜色,为不同颜色背景的图像选择最佳的预处理与分割方法;其次使用基于形状参数、尺度不变特征变换-快速近似最近邻查找(SIFT-FLANN)、方向梯度直方图-支持向量机(HOG-SVM)的多方法融合的藻类检测与识别算法,以提高识别的准确率和鲁棒性。对25种常见的藻类进行实验与对比,其中常见的10种藻类的平均检测成功率为96.07%,对鱼腥藻、针杆藻的识别准确率较高,分别达到了100.00%与99.50%,验证了该算法对常见藻类的检测与识别具有一定的鲁棒性。 |
| 关键词: 藻类检测与识别;SIFT-FLANN;HOG-SVM |
|
中图分类号: TP391
文献标识码: A
|
|
| Detection and Recognition of Common Freshwater Algae in Aquatic Environments Based on the Fusion of Multiple Method |
|
ZHANG Wenyi, ZHU Hong, NIE Fengxiang, ZHANG Jiajun
|
(Computer and Sof tware School, Dalian Neusof t University of Inf ormation, Dalian 116023, China)
zhangwenyi01@outlook.com; zhuhong@neusoft.edu.cn; a1748355368@outlook.com; 1282947838@qq.com
|
| Abstract: Aiming at the problems of low detection and recognition accuracy caused by the diverse morphology of the same type of algae, high similarity of algae, and the presence of bubbles and impurities in the collected images, this paper proposes a common algae detection and recognition algorithm in freshwater based on the fusion of multiple methods. Firstly, color information is used to identify the background color of algae images, allowing for the selection of optimal preprocessing and segmentation methods for images with different background colors. Secondly, a multimethod fusion algae detection and recognition algorithm based on shape parameters, Scale Invariant Feature Transform-Fast Library for Approximate Nearest Neighbors (SIFT-FLANN), and directional gradient Histogram-Support Vector Machine (HOG-SVM) is used to improve the accuracy and robustness of recognition. Through experiments and comparisons conducted on 25 common algae, an average detection success rate of the 10 common algae is 96.07% , and the recognition accuracy rate for Anabaena and Synedra is high, reaching 100.00% and 99.50% , respectively, verifying that the proposed algorithm has a certain degree of robustness for the detection and recognition of common algae. |
| Keywords: algae detection and recognition; SIFT-FLANN; HOG-SVM |
