The receiving array composed of single-photon avalanche diodes (SPADs) can effectively improve the sensitivity of the receiving end, which has important application value in the field of remote detection and imaging. For moving targets in the air, it is difficult to obtain stable target echoes due to the limited single-pixel acquisition time. In particular, in long-distance conditions, the laser energy attenuates significantly in the channel transmission process, which makes the number of echoes detected by the SPAD array in imaging lessen. As the detector does not fully accumulate the echoes, the difficulty of target recognition rises due to the lack of target image features in its imaging results. For the accurate detection of such targets, target recognition methods are required to make full use of limited echo information. Through the method of image reconstruction, the image quality can be optimized to a certain extent, so as to improve the system's ability to recognize the measured target. However, image reconstruction requires a long processing time, which is difficult to meet the real-time requirements of moving target monitoring. For timely and effective identification of moving targets in the air, the detection system should have the ability to quickly process image information.

Under the condition of array imaging with low resolution, few features, and serious noise interference, traditional image processing methods and contour processing methods can hardly ensure timeliness and accuracy due to a large amount of data and great time consumption. When the weak imaging result of the SPAD array is directly used for target recognition, it does not require high-quality reconstruction of the target shape and texture. Hence, it can effectively reduce the data requirements of image reconstruction and the complexity of algorithms and is of great significance for realizing real-time monitoring of long-distance moving targets in the air. For a common low-altitude aircraft, its deformation rate is far lower than its displacement change during the movement, and thus, it is not necessary to recognize the target synchronously during target tracking. Therefore, the following solution is proposed: the detection process can be divided into two parts, i.e., target tracking and target recognition. On the basis of target positioning and tracking in a single imaging frame, multiple imaging frames are used for target recognition to neutralize the contradiction between recognition effect and processing speed. Upon the above considerations, this paper proposes an optical flow method based on clustering analysis and optical flow features.

The method proposed in this paper can accomplish real-time tracking and recognition of moving targets in the air without any a priori information (Figs. 1-2). Considering the complexity of the moving mode of the airborne flying target, it is necessary to simulate three-dimensional motion information with a two-dimensional optical flow field. Since dimensions are reduced by the direct removal of depth information, the overlapping problem of multiple targets occurs. Therefore, this paper uses the projection method of aperture imaging to convert motion information to optical flow information (Figs. 3-5). To verify the effectiveness of the proposed method, this paper obtains more effective classification criteria through the statistics and analysis of the optical flow angle data of "low, slow, and small" targets and verifies the feature recognition results of optical flow angles according to the change in the value of the optical flow mode (Fig. 8). Upon the removal of the imaging frames with abnormal moduli, the experimental statistical results of the overall optical flow angle vector are consistent with the theoretical analysis results (Fig. 9). In target classification, this method uses the essential motion characteristics of the flying target, which is free from the interference of various types of shape camouflage and has a wide application scope.

Under the condition of array imaging with low resolution, few features, and serious noise interference, it is difficult to consider real-time detection and accurate target recognition by traditional target recognition methods due to the massive data to be processed and the huge time consumption. To alleviate the contradiction between timeliness and accuracy, this paper proposes an optical flow feature recognition method on the basis of the flight characteristics of different targets, which overcomes the recognition difficulty caused by poor array imaging effects. Due to limited time, this paper only conducts experiments and analysis on typical targets such as fixed wing UAVs, rotary wing UAVs, and birds. In the future, it is expected that the optical flow recognition method will be extended to more targets such as airships, balloons, and gliders to prove the universal applicability of this method in long-range aerial target detection. As the hardware processing capability is enhanced, the method of image feature recognition will have more advantages in real-time target detection, which should be the focus of future research.