Airborne LiDAR, as a precise and fast earth’s surface three-dimensional (3D) measuring method, has been widely used in the past decades. It provides a new approach for acquiring road information. By analyzing the characteristics of LiDAR datasets as well as that of the road in the datasets, a morphological method has been proposed to automatically extract the road from airborne LiDAR datasets. Firstly, ground points are segmented from raw LiDAR data by morphological operations. The key factor in this process is how to select the window sizes in different scale spaces, and setting the elevation threshold to prevent over-segmentation in each scale space. Secondly, candidate road points are segmented from the ground points, which are obtained from previous step, by intensity constraint, local point density and region area constraint, and so on. Thirdly, morphological opening operation and closing operation were used to process the candidate road points segmented from above steps. The opening operation may effectively filter the noise areas, and greatly maintain the road detail. The closing operation may fill the small holes within the road, connecting nearby roads, and smoothing the road boundary, without signification area change. The main road can be extracted from the raw airborne LiDAR points by previous three steps. Finally, the proposed method has been verified by LiDAR data which consists of complex road networks. The result shows that the proposed method can automatically extract road from airborne LiDAR data with higher efficiency and precision.
A set of Raman spectrum measurement system, essentially a Raman spectrometer, has been independently designed and accomplished by our research group. This system adopts tiled-grating structure, namely two 50mm × 50mm holographic gratings are tiled to form a big spectral grating. It not only improves the resolution but also reduces the cost. This article outlines the Raman spectroscopy system’s composition structure and performance parameters. Then corresponding resolutions of the instrument under different criterions are deduced through experiments and data fitting. The result shows that the system’s minimum resolution is up to 0.02nm, equivalent to 0.5cm-1 wavenumber under Rayleigh criterion; and it will be up to 0.007nm, equivalent to 0.19cm-1 wavenumber under Sparrow criterion. Then Raman spectra of CCl4 and alcohol have been obtained by the spectrometer, which agreed with the standard spectrum respectively very well. Finally, we measured the spectra of the alcohol solutions with different concentrations and extracted the intensity of characteristic peaks from smoothed spectra. Linear fitting between intensity of characteristic peaks and alcohol solution concentrations has been made. And the linear correlation coefficient is 0.96.
In order to accomplish recognition of the different edible oil we set up a laser induced fluorescence spectrum system in the laboratory based on Laser induced fluorescence spectrum technology, and then collect the fluorescence spectrum of different edible oil by using that system. Based on this, we set up a fluorescence spectrum database of different cooking oil. It is clear that there are three main peak position of different edible oil from fluorescence spectrum chart. Although the peak positions of all cooking oil were almost the same, the relative intensity of different edible oils was totally different. So it could easily accomplish that oil recognition could take advantage of the difference of relative intensity. Feature invariants were extracted from the spectrum data, which were chosen from the fluorescence spectrum database randomly, before distinguishing different cooking oil. Then back propagation (BP) neural network was established and trained by the chosen data from the spectrum database. On that basis real experiment data was identified by BP neural network. It was found that the overall recognition rate could reach as high as 83.2%. Experiments showed that the laser induced fluorescence spectrum of different cooking oil was very different from each other, which could be used to accomplish the oil recognition. Laser induced fluorescence spectrum technology, combined BP neural network，was fast, high sensitivity, non-contact, and high recognition rate. It could become a new technique to accomplish the edible oil recognition and quality detection.