3.2. Flight Experiment with Complex Terrain Observation Target
In order to verify the proposed spatial projection method of complex topographic data, we selected a small lake surrounded by bare soil on the campus of the Northeast Institute of Geography and Agroecology, Chinese Academy of Sciences located in Changchun, Jilin Province, China, as the experimental area. The experiment was carried out in winter when the lake was frozen, and there was 1~3 cm of snow (microwave can penetrate thin snow, so it had no effect on the experiment). The elevation of the lake was 194 m, the lowest in the region, and its edge gradually rose to 200 m. The experimental area was a rectangle of 270 m × 80 m. A total of five routes were designed, of which three flew west, and two flew east, with azimuths of 270° and 90°, respectively, and an incident angle of 55° for ground observation. The route length was about 210 m, and the route spacing was about 15 m.
The experimental results are shown in
Figure 14;
Figure 14a1–a3 shows the optical image and the flying routes, the digital surface model (DSM) data, and the ground object classification results of the study area, respectively. Using the data processing method in
Section 2.2, the TB swath data and the TB gridded data (resampling spatial resolution of 0.0001°) in the test area were obtained, as shown in
Figure 14b1–b2. The results show that the TB of the ice surface was relatively low, ranging from 195–215K, while that of the bare soil area was relatively high, exceeding 250K, and the middle area was a transition strip. The system could accurately measure the TB of different ground objects and provide the value of pure pixels.
A threshold of 220K was selected to classify the lake ice area and non-lake ice area, and the binarized raster data are shown in
Figure 14b3. Compared to the classification results of the optical image, the ice surface range identified by microwave was relatively small, which might have been caused by the following: (1) the selection of the threshold value would have affected the classification results. The pure lake ice pixels were selected, and the mixed pixels within transition bands were not included. (2) The ice thickness at the edge of the lake was relatively small, and the microwave could penetrate it to a certain degree, so the bare soil under the ice was detected. The classification results could roughly distinguish the boundary and were in good agreement with the actual boundary shape.
Meanwhile, the performances of the same set of original data for uncorrected, angle-corrected, and terrain-corrected data were compared to verify the proposed algorithm and method. The TB gridded data and binary classification results are shown in
Figure 15. The results show that, after the angle correction, the position of the projection point was obviously offset, which led to an obvious ‘in–out’ phenomenon of the lake ice area in the image, and this phenomenon significantly improved after the topographic correction.
Compared to the ground-based measurement, the incidence angle would be changed by the flight attitude of the UAV, which would affect the projection position of the beam center and the range of the FOV. The incidence angle on the ground was 55° and 50° after the modification of the UAV attitude. When the flight altitude was 30 m, the central projection point needed to be corrected 7 m backward along the observation direction. On this basis, since the altitude of the survey area was lower than the takeoff altitude, according to the previous analysis, the central projection point needed to be corrected forward along the observation direction, and the amount of correction depended on the altitude difference between the survey area and the takeoff position.
The corrected offsets, or the distances between the uncorrected points and corrected points using the data processing method proposed in sub-section “The Projection Process”, were measured, and the altitudes of the projection points were extracted. Then, the theoretical quantities of the offsets could be calculated using the altitude differences with
where h and h′ are the flying height and altitude difference, respectively, and θ and θ′ are the observation angle before and after correction, respectively. The theoretical quantities and corrected offsets of the proposed method were compared, and the result is shown in
Figure 16. The altitude difference between the measurement target and the takeoff position is shown in different colors. The calculated theoretical value and the corrected value based on the actual angle and terrain were very close to the 1:1 line, and the R
2 was 0.87, which verifies the effectiveness of the angle and terrain correction method proposed.
Interestingly, there was a certain degree of similarity between the uncorrected results and the corrected results in determining the pure pixels of the lake ice. When not corrected, the forward projection distance of the central beam along the observation direction was , while the altitude difference between the lake ice and the takeoff position was 6 m, so on the altitude profile of the lake ice, the central beam forward projection distance along the observation direction was . It was only a numerical coincidence that the values were very close to the uncorrected ones, which led to similar results in the lake ice region. For the projection points in the bare soil area at other altitudes, the correction effect was obvious, approximately 2–7 m (based on the altitude difference). However, this result is not obvious in the raster data with a spatial resolution of 0.0001°.
In addition, we only calculated the incidence angle and FOV corrected according to the actual terrain but did not correct the TB according to the corrected incidence angle, which might be the work we may pay attention to later.