Research On Aerial Remote Sensing System Based Small UAV And Its Image Processing

Recently, with the rapid development of social economy, the monitoring accuracy requirements need to improve in terms of agriculture, resources, environment, disaster mitigation, surveying and mapping, and large-scale infrastructure construction, sharply. Therefore, the study of remote sensing images based on hyperspectral resolution will play a crucial role in the monitoring of the surface. Since 1980 s, hyperspectral remote sensing technology has been a popular field of earth observation, and it’s a technology based on the principles of spectroscopy, which obtains the interesting information of ground cover by combining with a large amount of narrow-band electromagnetic wave. However, the traditional hyperspectral remote sensing images are always photographed by satellite, which has lots of problems such as weak information timeliness, relatively long cycle, low spectral resolution and easily affected by weather conditions, etc.In recent years, with the rapid development of auto-control technology, computer science technology and sensor technology, Unmanned Aerial Vehicle(UAV) aerospace technology develops rapidly. Its task has converted from testing target drone to aerial reconnaissance, and its application is gradually extending from military fields to civil fields. Compared with the traditional aerospace remote sensing technology, UAV Hyperspectral Remote Sensing has many advantages such as low development costs, short development cycle and low operating cost. Therefore, the mature development of UAV aerospace technology provides strong support for obtaining hyperspectral resolution and high-efficient remote sensing images, and it will be an inevitable development trend of hyperspectral remote sensing technology in the future for the serving of national economic construction by obtaining Hyperspectral Remote Sensing Images using UAVs.In the UAV hyperspectral remote sensing system, it mainly concludes the design of UAV flight control system, as well as the feature extraction of hyperspectral remote sensing image in the follow-up process. However, the core technology of UAV flight control system is the navigational system. Therefore, this paper will research in the terms of navigation system design and feature extraction algorithm of hyperspectral remote sensing image.① This paper presents a nonlinear navigation tracking control method for UAV, which obtains the UAV’s location and flight direction in real time through GPS module and performs navigation tracking control in a fixed navigation control cycle. When it reaches its navigation control cycle, the nonlinear navigation tracking control method performs navigation tracking control. According to different situations, we caculate the navigation angle by the pre-set route and the UAV’s current location. When UAV is far away from the pre-set route, the angles between the flight and the route are relatively big, resulting in a large value of centripetal acceleration or transverse acceleration. Therefore, it will cause a relatively big navigation angle, which makes the UAV approach the route closely and sharply. On the contrary, the UAV will approach the route with a small navigation angles. With this nonlinear navigation tracking control method, the UAV can change the navigation angle q according to the specific flight status and make q more accurate in a relatively complex task. Navigation algorithm parameters can be adjusted dynamically according to the UAV’s flight speed. The result of the experiment by the autonomous designed UAV flight control system platform shows that the nonlinear navigation tracking control method can meet the demand of practical projects.②This paper proposes a new semi-supervised manifold learning method called Semi-supervised Sparse Discriminant Embedding(SSDE). The algorithm combines the advantages of manifold structure among classes and sparsity, which not only preserves the sparse reconstruction relationship and sparsity, but gets the intrinsic manifold structure of data by taking use of a few labeled training samples and a large number of unlabeled training samples, to extract discriminant feature of data and improve classification accuracy. The classification experiments in Washington DC Mall and Indian Pine data set show that the method is a more effective way to find the internal structure of data in high dimensional space. Compared to other methods, SSDE obviously improves the classification performance. Take the random selected 8 training samples with classification labels as an example, the classification precision of SSDE respectively reach 77.36% in Indian Pine and 97.85% in Washington DC Mall data set.For the UAV hyperspectral remote sensing navigation system and the follow-up image processing, we design and verify the effectiveness of the algorithm with experiment. Finally, for the further optimization of the UAV aerial remote sensing system, some feasible suggestions are put forward.