Classifying land cover based on calibrated full-waveform airborne light detection and ranging data

The capability of the parameters derived from waveform data in discriminating objects is assessed and the effect of the relative calibration of full-waveform data in discriminating land-cover classes is evaluated. Firstly, a non-linear least-squares method with the Levenberg-Marquardt algorithm is used to fit the return waveforms by a Gaussian function. Gaussian amplitude, standard deviation, and energy are extracted. Secondly, a relative calibration method using the range between the sensor and the target based on a radar equation is applied to calibrate amplitude and energy. The change in transmit pulse energy is also considered in this process. A support vector machine classifier is used to distinguish the study area into non-vegetated area （including roads, buildings, and vacant lots）, grassland, needle-leaf forests, and broad- leaf forests. The overall classification accuracy ranges from 79.33% to 87.6%, with the combination of the two groups of the three studied parameters. Calibrated data classification accuracy is improved from 1.20% to 6.44%, thus resulting in better forest type discrimination. The result demonstrates that the parameters extracted from the waveforms can be applied effectively in identifying objects and that relative calibrated data can improve overall classification accuracy.