Classification of full waveform data for monochromatic airborne LiDAR bathymetry based on waveform morphological features

  Objective  Monochromatic airborne LiDAR bathymetry becomes considerably favorable for topography and geomorphology detection over coastal area by means of its low cost, low load and high sampling rate. However, addressing the limitation of single wavelength to realize the accurate division of full waveform data independently from auxiliary sensor becomes the critical part of coordinate calculation. Given the existing literatures, there is a lack of systematic evaluation analysis and general conclusions for waveform classification contraposing to full waveform morphological features.

  Methods  In view of the latest development of waveform features extraction, refined waveform categories (anomalies, over-fitted, land, sea surface and bathymetry waveforms), 24-dimensional waveform features are designed and calculated upon systematic analysis on morphological characteristics of different waveforms, and then their classification performance and optimal feature combination are evaluated and quantitatively analyzed utilizing random forest feature selection and classification model.

  Results and Discussions   The results proved that the combination of 6-dimensional features (Fig.8-11), including deviation of amplitude between two adjacent points and oscillating main frequency, is the most effective in classifying five waveforms, with an overall classification accuracy of 98.55% and a Kappa coefficient of 0.982 0 (Fig.9-10, Fig.12, Tab.1). To verify the universality of the features, an additional experimental area was selected for validation and the overall accuracy of water and land classification was 96.81% (Fig.13).

  Conclusions  To accurately identify waveforms, a systematic analysis was conducted to determine the morphological differences between different types of waveforms, and 24-dimensional feature parameters were extracted. After the optimal feature combination and classification performance evaluation, it was found that the 6-dimensional features of oscillating main frequency f, ratio of peak R_p, deviation of amplitude between two adjacent points ∆A, maximum intensity Wf-I, decay constant a, and first echo peak Fpk were highly effective in distinguishing these five types of waveforms, where 100% of the anomalies and over-fitted waveforms were extracted, strongly confirming the relevance and validity of the morphological features. After replacing the experimental area, the accuracy of the water and land classification reached 96.81%, proving that the features and methods used were adaptable and generalizable, and could meet the production requirements. The 2% decrease in waveform classification accuracy after changing the study area is mainly due to the varying equipment parameter settings and coverage feature categories in different experimental areas. Limited sample selection further compounds this issue. To maintain accuracy, the sample data can be appropriately supplemented according to the actual situation in the experimental area. Although the waveform morphology has been studied thoroughly, additional experimental evidence is necessary to ascertain the impact of intrinsic factors such as signal reception systems on the echoes. To this end, future research will focus on expanding the study's data and signal detection.