Identification Algorithm of Green Algae Using Airborne Hyperspectral and Machine Learning Method
SUN Lin1, BI Wei-hong1, LIU Tong1, WU Jia-qing1, ZHANG Bao-jun1, FU Guang-wei1, JIN Wa1, WANG Bing2, FU Xing-hu1*
1. School of Information Science and Engineering, The Key Laboratory for Special Fiber and Fiber Sensor of Hebei Province, Yanshan University, Qinhuangdao 066004, China
2. Qinhuangdao Hongyan Photoelectric Technology Co., Ltd., Qinhuangdao 066004, China
Abstract:The green tide is a kind of algal bloom phenomenon formed by the growth and aggregation of Marine macroalgae, which seriously affects the coastal ecological environment. Accurate monitoring of the coverage area of green tide is of great significance for preventing, monitoring and managing green tide disasters. The use of spectral methods for remote sensing monitoring has the advantages of non-contact, low cost and small loss, among which airborne hyperspectral remote sensing has a wide range of application prospects in the Marine field due to its advantages of high spectral and spatial resolution and more imaging channels. This study used DJI M300 RTK equipped with 410 Shark hyperspectral imaging system to collect data from the green algae outbreak area in The Dream of Bay, Qinhuangdao City. The collected spectral data were preprocessed to extract the spectral features of different ground objects. Based on the features, a spectral feature dataset with a capacity of 30 000 was constructed. The dataset was randomly divided into the training set and a test set, in which the training set accounted for 75% and the test set accounted for 25%. Five machine learning algorithms established the hyperspectral green tide inversion model, including Decision Tree, Random Forest, SVM, K-Nearest Neighbor and three-input voting classifier. The green algae coverage area in the green tide outbreak area was calculated by ground resolution cell (GRC) based on an airborne hyperspectral imaging system, and the classification accuracy of the inversion model was tested based on the in-dataset accuracy, Kappa coefficient and the preset standard area error verification method. The experimental results show that time can be saved by band selection first when the dichotomy of green algae pixels and other earth objects is performed on hyperspectral data and big data prediction is performed using the proposed classifier. The classification accuracy of the hyperspectral data can be effectively improved by logarithmic processing to enhance the differences between the spectra and then constructing the classifier model. The inversion accuracy of the hyperspectral green tide inversion model based on the three-input voting classifier of Random Forest, SVM and K-Nearest Neighbor is 98.95% in the dataset, and the Kappa coefficient is 0.978 9. The classification error obtained by the prespecified standard area error verification method is 6.06%. Applied through the experimental area of hyperspectral image prediction, it proved that the model in the prediction of big data still keeps high accuracy, and for the mixed pixels underwater algae pixels can also define, show that the method is feasible and superiorin the field of green tide remote sensing monitoring, in the field of green tide area monitoring has universality, has extensive application prospect in the field of marine monitoring.
孙 琳,毕卫红,刘 桐,武家晴,张保军,付广伟,金 娃,王 兵,付兴虎. 应用机载高光谱与机器学习法的绿藻识别算法研究[J]. 光谱学与光谱分析, 2023, 43(11): 3637-3643.
SUN Lin, BI Wei-hong, LIU Tong, WU Jia-qing, ZHANG Bao-jun, FU Guang-wei, JIN Wa, WANG Bing, FU Xing-hu. Identification Algorithm of Green Algae Using Airborne Hyperspectral and Machine Learning Method. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2023, 43(11): 3637-3643.
[1] BAI Yu, ZHAO Liang, LIU Jing-zhou(白 雨, 赵 亮, 刘境舟). Haiyang Xuebao (海洋学报), 2019, 41(8): 97.
[2] Liu D Y, John K K, He P M, et al. Estuarine, Coastal and Shelf Science, 2013, 129: 2.
[3] Ye N H, Zhang X W, Mao Y Z, et al. Ecological Research, 2011, 26(3): 477.
[4] FAN Shi-liang, FU Ming-zhu, LI Yan, et al(范士亮, 傅明珠, 李 艳, 等). Haiyang Xuebao(海洋学报), 2012, 34(6): 187.
[5] CHEN Ying, SUN De-yong, ZHANG Hai-long, et al(陈 莹, 孙德勇, 张海龙, 等). Acta Optica Sinica(光学学报), 2020, 40(3): 0301001.
[6] LI Zhao-xin, QIU Zhong-feng, LI Xu-wen, et al(李兆鑫, 丘仲锋, 李旭文, 等). Environmental Monitoring and Forewarning(环境监控与预警), 2019, 11(5): 46.
[7] XING Qian-guo, ZHENG Xiang-yang, SHI Ping, et al(邢前国, 郑向阳, 施 平, 等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2011, 31(6): 1644.
[8] Wang M Q, Hu C M. Remote Sensing of Environment, 2016, 183: 350.
[9] LIU Jin-chao, LIU Jian-qiang, DING Jing, et al(刘锦超, 刘建强, 丁 静, 等). Haiyang Xuebao(海洋学报), 2022, 44(5): 1.
[10] Xing Q G, Hu C M. Remote Sensing of Environment, 2016, 178: 113.
[11] ZHANG Hai-long, SUN De-yong, LI Jun-sheng, et al(张海龙, 孙德勇, 李俊生, 等). Acta Optica Sinica(光学学报), 2016, 36(6): 0601004.
[12] Jannoura R, Brinkmann K, Uteau D, et al. Biosystems Engineering, 2015, 129: 341.
[13] XU Fu-xiang, GAO Zhi-qiang, SHANG Wei-tao, et al(徐福祥, 高志强, 尚伟涛, 等). Oceanologia Et Limnologia Sinica(海洋与湖沼), 2018, 49(5): 1061.
[14] LI Dong-xue, GAO Zhi-qiang, SHANG Wei-tao, et al(李冬雪, 高志强, 尚伟涛, 等). Marine Environmental Science(海洋环境科学), 2020, 39(3): 438.
[15] ZHANG Chun-min, MU Ting-kui, YAN Ting-yu, et al(张淳民, 穆廷魁, 颜廷昱, 等). Spacecraft Recovery & Remote Sensing(航天返回与遥感), 2018, 39(3): 104.
[16] MIAO Chun-li, FU Shuai, LIU Jie, et al(苗春丽, 伏 帅, 刘 洁, 等). Pratacultural Science(草业科学), 2022, 39(10): 1992.
[17] Ministry of Natural Resources, People's Republic of China(中华人民共和国自然资源部). HY/T 0331—2022, Technical Specifications for Ecological Survey and Monitor on Green Tide(绿潮生态调查与监测技术规范), 2002.
[18] LI Tong-ji(李铜基). China's Offshore Oceans-Ocean Optical Characteristics and Remote Sensing(中国近海海洋:海洋光学特性与遥感). Beijing: China Ocean Press(北京:海洋出版社),2012: 273.