Abstract

Unmanned Aerial Vehicle (UAV) may provide us super resolution data, however, they are often captured with artefacts and distortion. To investigate impact of distortion on the coral reef information, UAV surveys were deployed using multispectral sensor over two reefs in Bidong Island (Malaysia). Band-specific analysis of distortion revealed five different types of distorted images from the acquisition. This study optimized screening distorted images by comparing the seven distortion correction approaches and validated coral classification maps based on machine learning algorithms [support vector machine (SVM), random forest (RF) and artificial neural network (ANN)]. Results indicate that the screening the green band (b2) alone or the blue band (b1) combined with b2 of UAV data and SVM capable of generating the best coral classification maps, with an overall accuracy of 7–17% improved compared to that of orthomosaic without distortion correction. The proposed distortion correction method can be applied to similar coral environments.

• Highlights

• Five different types of artefacts and distortions found in UAV data.

• A optimized screening approach suggested to minimized image distortion.

• Distortion corrected data and SVM algorithm performed the best in coral habitat mapping.

• An overall accuracy of 7–17% improved compared to that of distortion uncorrected maps.

• A new photogrammetric contribution to the existing automatic orthorectification techniques.

摘要

无人驾驶飞行器 (UAV) 可以为我们提供超分辨率数据，但是，它们通常被捕获时带有伪影和失真。为了调查失真对珊瑚礁信息的影响，在 Bidong 岛（马来西亚）的两个珊瑚礁上使用多光谱传感器部署了无人机调查。失真的特定波段分析揭示了来自采集的五种不同类型的失真图像。本研究通过比较七种失真校正方法和基于机器学习算法 [支持向量机 (SVM)、随机森林 (RF) 和人工神经网络 (ANN)] 验证的珊瑚分类图来优化筛选失真图像。结果表明，单独筛选绿带（b2）或蓝带（b1）结合无人机数据和支持向量机的b2能够生成最佳珊瑚分类图，与没有失真校正的正射镶嵌相比，整体精度提高了 7-17%。所提出的失真校正方法可以应用于类似的珊瑚环境。

• 强调

• 在无人机数据中发现五种不同类型的伪影和失真。

• 建议使用优化的筛选方法来最小化图像失真。

• 失真校正数据和 SVM 算法在珊瑚栖息地测绘中表现最好。

• 与失真未校正的地图相比，整体准确度提高了 7-17%。

• 对现有自动正射校正技术的新摄影测量贡献。