Research On Ground Target Recognition Algorithm For LADAR Range Imagery

LADAR(Laser imaging radar) range imagery can reflect three-dimensional features of the target, therefore, it’s very suitable for the identification of ground targets from missile and has many potential and promising applications. At present, there are still many problems to be solved in the recognition of range imagery. The range imagery will be severely disrupted when the missile-borne miniature LADAR detects ground targets, because of the poor working condition and the complexity of the ground information. Meanwhile, the features of targets will be changed a lot, since the angle of LADAR sight is usually small and continually changing. Therefore, the noise suppression of LADAR range imagery, the ground extraction in a small angle of LADAR sight, the target classification and extraction, and the target feature extraction and recognition must be study in-depth.In this paper, the ground target recognition algorithms of LADAR range imagery are studied, which include the noise suppression, the ground extraction, the target extraction and the target recognition.First, the noise suppression algorithm of LADAR range imagery is studied. The simulation model of noisy LADAR range imagery is established by analyzing the mechanism and the distribution character of the noise in LADAR range imagery. And it is verified by means of median filter, and it is also proved that the performance of the noise suppression can be evaluated accurately by analyzing the target component in range imagery respectively. An improved donut filter algorithm to suppress the noise is proposed in which the correction parameters are adjusted according to the size of details. And a noise suppression algorithm based on the statistical distribution is proposed, which takes the distribution character of the range abnormal noise into account. The experimental results show that both of the proposed algorithms can suppress noise effectively and protect details well.Secondly, the ground extraction algorithm of ground target range imagery is studied. A reliable ground point extraction algorithm based on the ground segment identification in scan lines is proposed, which includes the noise suppression, the demarcation point extraction, the segment merger and the ground segment identification in the scan line. The data fluctuation in the scan line is suppressed effectively by the adaptive adjustment of the window size of the median filter. And the demarcation points are extracted by the neighborhood direction vector dot product of each point in the scan line. The different segments in the same plane are merged by the analysis of the belongings of the line that segments. Then a ground segment identification algorithm is proposed by analyzing the ground features in the scan line. The experimental results show that the algorithm is not affected by the angle of sight, and the ground points extracted are highly reliable and widely shrouded.Thirdly, the target classification and extraction algorithm in range imagery is studied, which include the altitude calculation, the ground segmentation, the target classification, the obscured target compensation and the target extraction. The altitude of every point in range imagery is obtained by calculating the normal vector of reliable ground points. Then the ground and the targets are segregated accurately by region growing, and the classification of different target in range imagery is extracted by three-dimensional clustering. It is pointed that the obscured targets can be extracted as well by searching the edge, and the compensation strategy is analyzed. The rapid extraction of the probable targets are also realized by the adaptive adjustment of the morphology structure element size. The experimental results show that the target classification and extraction algorithm can segregate the ground and the targets accurately, extract probable targets rapidly, remove false targets effectively, and handle obscured targets correctly.Finally, the target features extraction and recognition in range imagery are studied. The target altitude feature is extracted by the analysis of altitude probability density distribution. And the length and width is extracted by rectangular estimating. The image feature in specific projection directions are extracted as well by calibrating the target attitude. The target recognition algorithm based on artificial neural network is proposed according to the invariant target features extracted. The experimental results show that the invariant target features extracted are highly reliable with less computation, while the recognition algorithm can make full use of target priori knowledge with high efficiency, strong fault tolerance and good recognition capability.