利用Sentinel-2图像和机器学习模型检测海洋漂浮塑料

版权申诉

5星 · 超过95%的资源 142 浏览量更新于2024-07-06 收藏 3.32MB PDF 举报

"利用Sentinel-2图像和机器学习模型检测海洋漂浮塑料" 本研究论文旨在探索利用Sentinel-2图像和机器学习模型检测海洋漂浮塑料的可能性。海洋漂浮塑料污染是当今世界面临的一大环境问题，对海洋生态系统和生物多样性造成严重威胁。遥感技术和机器学习算法可以作为一种成本-effective解决方案，用于检测大型塑料污染。该研究使用了Sentinel-2卫星图像和两种机器学习模型，即支持向量机（SVM）和随机森林（RF），对四个地区的海洋漂浮塑料进行检测和分类：Mytilene（希腊）、Limassol（塞浦路斯）、Calabria（意大利）和Beirut（黎巴嫩）。研究结果表明，机器学习模型可以有效地检测和分类海洋漂浮塑料。 Sentinel-2卫星图像是一种开放的遥感数据源，提供了高分辨率的图像，可以用于检测海洋漂浮塑料。机器学习模型可以对图像进行分析和分类，从而检测出海洋漂浮塑料。研究结果表明，SVM和RF模型都可以达到良好的检测效果，但RF模型在检测海洋漂浮塑料方面具有更好的性能。该研究结果为检测海洋漂浮塑料提供了一种新的解决方案，具有成本-effective和环境友好的特点。同时，该研究也表明了机器学习模型在环境监测和污染检测方面的应用前景。知识点： 1. 海洋漂浮塑料污染是当今世界面临的一大环境问题，对海洋生态系统和生物多样性造成严重威胁。 2. 遥感技术和机器学习算法可以作为一种成本-effective解决方案，用于检测大型塑料污染。 3. Sentinel-2卫星图像是一种开放的遥感数据源，提供了高分辨率的图像，可以用于检测海洋漂浮塑料。 4. 机器学习模型可以对图像进行分析和分类，从而检测出海洋漂浮塑料。 5. 支持向量机（SVM）和随机森林（RF）都是常用的机器学习模型，用于检测和分类海洋漂浮塑料。 6. 机器学习模型可以用于环境监测和污染检测方面，具有广泛的应用前景。 7. 海洋漂浮塑料检测需要多种数据源和技术的结合，包括遥感技术、机器学习算法和现场调查等。 8. 机器学习模型在检测海洋漂浮塑料方面具有高度的准确性和可靠性。该研究论文探索了利用Sentinel-2图像和机器学习模型检测海洋漂浮塑料的可能性，为检测海洋漂浮塑料提供了一种新的解决方案。同时，该研究也表明了机器学习模型在环境监测和污染检测方面的应用前景。

used for validating the results. Additionally, for classifying the individual Sentinel scene

retrieved for Mytilene and Limassol, a total of 12 (for 7 June 2018), 8 (for 18 April 2019), 8

(for 3 May 2019), 12 (for 18 May 2019), 8 (for 28 May 2019), and 6 (for 12 December 2018)

samples were used to label the plastic class. For water class, a total of 68 (for 7 June 2018), 72

(for 18 April 2019), 72 (for 3 May 2019), 67 (for 18 May 2019), 62 (for 28 May 2019), and 43

(for 12 December 2018) samples were used for classification.

To test the model performances and assess the potential application of the developed

model on real-world data, all the 54 in-situ geocoded plastic data was used to train the ML

models. To handle the class imbalance problem, an equal number of water samples (58) were

collected from Mytilene and Limassol, and the same was used to label the water feature. The

model that performed most accurately on ground data (collected from Mytilene and Limassol)

was considered for predicting floating plastics in the testing sites (Beirut and Calabria). A total

of 42 and 33 samples were collected from Calabria and Beirut for labelling the plastic feature

for modelling. For the water class, a total of 82 and 53 samples were collected. The details of

training and testing data used in the ML models are provided in Fig. 1 and Fig. 2.

FDI and band reflectance in NIR and Red Edge 2 was used to discriminate the plastic

pixels from Sentinel true colour (RGB) image. A true-colour image was prepared for all four

locations for the mentioned dates. A visual inspection was carried out to detect the target plastic

pixels and verify the difference between the plastic pixels' spectral responses and the

surrounding water pixels. Followed by, the spectral responses of plastic pixels in the NIR and

Red spectral region were checked to analyse the deviation of the spectral signal emitted from

plastics. Using the geocoded (Latitude and Longitude) information of the targeted plastic pixels

available from Topouzelis et al. (2020) for Mytilene, Greece and Themistocleous et al. (2020)

for Limassol, Cyprus, the surface reflectance values of the 4 spectral bands, i.e. Red, Red

剩余44页未读，继续阅读

易小侠

粉丝: 6611
资源: 9万+

利用Sentinel-2图像和机器学习模型检测海洋漂浮塑料

基于Sentinel-2多光谱数据和机器学习算法的冬小麦LAI遥感估算.pdf

sentinel-transport-simple-http-1.8.0-API文档-中文版.zip

sentinel2-cloud-detector:适用于Python的Sentinel-2图像的Sentinel Hub Cloud Detector

DSen2:Sentinel-2图像的超分辨率

Sentinel-2：处理来自Sentinel-2卫星的多光谱图像

基于半监督学习的Sentinel-1 SAR图像洪水分割_Flood Segmentation on Sentinel-1 SA

sentinel:用于分析Sentinel-12图像的Python脚本

6s模型大气校正Python版本，支持GF1、2、Landsat-8、Sentinel-2等图像_大气校正.zip

Chlorophyll-a_Detection_Algorithms:利用叶绿素-a和光栅图像提取有害藻华检测值

颜色分类leetcode-my-Sentinel-2-notes:my-Sentinel-2-notes

最新资源