Cat-eye target detection active imaging system and method based on intelligent vision

Abstract

The present application relates to the field of photoelectric detection and intelligent identification, in particular to a cat-eye target detection active imaging system and method based on intelligent vision, comprising: an active illumination module for outputting a pulse laser with adjustable parameters, actively illuminating a target scene, and exciting a cat-eye effect backscatter echo of an optical device in the target scene; a coaxial imaging acquisition module for receiving the backscatter echo of the target scene and collecting information; an intelligent vision processing module for intelligently processing the received image data, outputting target information and scene perception data; an adaptive control module for constructing a closed-loop control link according to the scene perception data and the target information; a display and storage module for realizing human-computer interaction, displaying and storing real-time imaging pictures and target information. Thus, the problems of high false alarm rate, limited detection distance, poor environmental adaptability, and inability to realize target fine classification and continuous tracking in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of photoelectric detection and intelligent recognition technology, specifically to an active imaging system and method for detecting cat-eye targets based on intelligent vision. Background Technology

[0002] The cat's-eye effect refers to the phenomenon where, when a laser beam strikes an optical lens (such as a camera lens, telescope, sniper rifle scope, drone electro-optical pod, or lens of a spy camera), a strong back reflection occurs due to the photosensitive element or reticle on the focal plane of the optical system. The intensity of the reflected light is 2 to 4 orders of magnitude higher than that of ordinary diffuse reflection, and the reflected light returns along the original incident light path, exhibiting extremely strong directionality. Active detection technology based on the cat's-eye effect is a core technology in the field of concealed optical target detection and has wide-ranging applications in national defense, public security, and counter-espionage reconnaissance.

[0003] Currently, several active detection systems based on the cat's eye effect have been deployed both domestically and internationally. However, due to limitations in imaging mechanisms and signal processing methods, existing systems still suffer from the following core technical shortcomings in practical applications: (1) Existing systems mostly use traditional image processing algorithms such as threshold segmentation and morphological processing, relying on only one or a few features such as spot brightness, size, and circularity to identify cat-eye targets. In real complex background environments, non-cooperative targets such as specular reflection, metal component reflection, glass curtain walls, water reflections, and vegetation highlights are very likely to produce bright spot echoes similar to the cat-eye effect. Traditional algorithms cannot effectively distinguish between real cat-eye targets and the above-mentioned false interference sources, resulting in a high false alarm rate of the system, which is difficult to meet the practical needs of real battlefield or security scenarios; (2) Most existing systems adopt a transceiver split optical path design, that is, the illumination optical path and the imaging optical path are separated. When detecting at long distances, this design causes the matching error between the laser divergence angle and the imaging field of view to increase with distance, and the laser illumination area and the imaging field of view cannot be accurately aligned, resulting in detection blind spots or insufficient edge illumination; at the same time, the off-axis design leads to a decrease in echo energy utilization, and the system signal-to-noise ratio deteriorates rapidly with distance, which seriously limits the maximum detection distance. (3) The laser illumination parameters (power, repetition rate, divergence angle, etc.) and imaging acquisition parameters (exposure time, gain, frame rate, etc.) are mostly factory-set or manually preset, and cannot be adjusted in real time according to dynamic changes such as ambient light intensity, target distance, and weather conditions (rain, fog, haze). In complex scenarios such as strong light, backlight, low illumination, or rain and fog, the image may be overexposed, underexposed, or the signal-to-noise ratio may drop sharply. The system's detection robustness is significantly insufficient, making it difficult to achieve stable operation in all weather conditions.

[0004] (4) Most existing systems can only detect the presence or absence of cat-eye targets and output simple alarm signals. They cannot classify the types of cat-eye targets in a refined manner (e.g., distinguishing between camera lenses, UAV electro-optical pods, sniper scopes, vehicle-mounted optical equipment, etc.). At the same time, existing systems lack the ability to continuously lock onto and track moving targets, and cannot provide accurate and continuous data support for subsequent threat assessment, target disposal, or fire guidance.

[0005] In summary, current active target detection technology still has significant shortcomings in areas such as false alarm suppression, long-range detection, environmental adaptation, and precise target identification and tracking. There is an urgent need for a system-level solution that can simultaneously address these issues. Summary of the Invention

[0006] This application provides an active imaging system and method for cat eye target detection based on intelligent vision, in order to solve the problems of high false alarm rate, limited detection distance, poor environmental adaptability, and inability to achieve fine target classification and continuous tracking in the prior art.

[0007] The first aspect of this application provides an active imaging system for cat-eye target detection based on intelligent vision, comprising: an active illumination module, a coaxial imaging acquisition module, an intelligent vision processing module, an adaptive control module, and a display, control, and storage module; wherein, The active illumination module is used to output pulsed laser with adaptively adjustable parameters to actively illuminate the target scene and stimulate the cat's eye effect back reflection echo of the optical device in the target scene. The coaxial imaging acquisition module and the active illumination module adopt a strictly coaxial optical path design for receiving and transmitting back the back reflection echo of the target scene, completing photoelectric conversion and image acquisition, and transmitting the acquired image data to the intelligent vision processing module. The intelligent vision processing module is connected to the coaxial imaging acquisition module, the adaptive control module, and the display, control and storage module respectively. It is used to perform intelligent vision processing on the received image data, complete the identification, positioning and tracking of the cat eye target, and output target information and scene perception data. The adaptive control module is connected to the active lighting module, the coaxial imaging acquisition module, and the intelligent vision processing module respectively. It is used to adaptively adjust the laser output parameters of the active lighting module and the imaging acquisition parameters of the coaxial imaging acquisition module according to the scene perception data and target information, and construct a closed-loop control link. The display, control and storage module is used to realize human-computer interaction, display of real-time imaging images and target information, and storage of raw images, target data and system logs.

[0008] Preferably, the active illumination module includes a multi-band laser source, a beam shaping unit, an acousto-optic modulator, and a laser driving circuit; wherein, The multi-band laser source uses a near-infrared pulsed laser, with an output wavelength covering at least one of 808nm, 940nm, and 1064nm, to achieve concealed active illumination. The beam shaping unit includes a collimating lens group and an adjustable beam expander, which are used to collimate and expand the laser beam to achieve adaptive adjustment of the laser divergence angle within the range of 0.1 mrad to 5 mrad. The acousto-optic modulator is used to modulate the laser pulse and output a nanosecond-level narrow pulse laser with an adjustable pulse repetition rate of 10Hz~10kHz. The laser driving circuit is connected to the adaptive control module and is used to receive control commands and adjust the laser's output power, repetition rate, pulse width, and divergence angle.

[0009] Preferably, the coaxial imaging acquisition module includes a polarizing beam splitter, a quarter-wave plate, an imaging objective lens group, a narrow-band filter, a planar photodetector, and an image acquisition circuit; wherein, The polarizing beam splitter, in conjunction with the quarter-wave plate, forms a coaxial optical path for transmitting and receiving: the laser output from the active illumination module is reflected by the polarizing beam splitter and then emitted to the target scene through the imaging objective lens group; the back reflection echo from the target scene returns along the original optical path, and after passing through the imaging objective lens group, the polarizing beam splitter, and the quarter-wave plate, it is transmitted and enters the narrowband filter. The center wavelength of the narrowband filter is matched with the laser wavelength of the active illumination module, and the half-width at half-maximum (WHM) is ≤10nm, which is used to filter out stray light from the ambient background. The imaging objective lens group adopts a telephoto zoom objective lens with an adjustable focal length range of 50mm to 500mm and an adjustable field of view range of 0.1° to 5°. The area array photodetector uses a near-infrared enhanced CMOS / CCD / InGaAs detector, supports global exposure, and has an adjustable frame rate range of 1fps to 60fps. The image acquisition circuit is used to perform analog-to-digital conversion and preprocessing of the image, and transmit the acquired image data to the intelligent vision processing module.

[0010] Preferably, the intelligent vision processing module adopts an FPGA+ARM heterogeneous computing architecture or an edge GPU computing module, including an image preprocessing unit, a candidate region extraction unit, a deep learning feature extraction and classification unit, a target localization and ranging unit, and a multi-frame correlation tracking unit; wherein, The image preprocessing unit is used to sequentially perform blind pixel correction, non-uniformity correction, adaptive denoising, contrast enhancement, and background difference suppression on the received raw image. The candidate region extraction unit is used to integrate adaptive threshold segmentation connected component analysis with a lightweight target detection network to quickly filter out candidate regions of suspected cat-eye targets and generate candidate target boxes. The deep learning feature extraction and classification unit has a built-in lightweight CNN classification network, which is used to extract features from image slices of candidate regions, distinguish between real cat eye targets and false interference sources, and output the category and corresponding confidence of the cat eye target. The target positioning and ranging unit is used to calculate the pixel coordinates of the target in the image based on the identified cat eye target, obtain the azimuth and pitch angles of the target by combining the imaging intra-participant calibration parameters, calculate the straight-line distance of the target based on the laser time-of-flight (TOF) principle, and output the three-dimensional position information of the target. The multi-frame correlation tracking unit uses an improved DeepSORT multi-target tracking algorithm, combined with Kalman filtering, to perform feature matching and trajectory correlation on cat-eye targets in consecutive frames, and outputs the target's motion trajectory.

[0011] Preferably, the training dataset of the lightweight CNN classification and recognition network includes real cat eye target samples with different distances, different lighting, and different backgrounds, as well as various fake interference samples. The dataset is expanded by data augmentation methods such as rotation, scaling, noise addition, and brightness adjustment. The lightweight CNN classification and recognition network achieves an accuracy rate of ≥99% and a false alarm rate of ≤0.1% for identifying cat-eye targets. The lightweight CNN classification network can identify cat-eye target categories including at least one of camera lenses, telescopes, sniper scopes, drone optoelectronic pods, and vehicle-mounted optical equipment.

[0012] Preferably, the adaptive control module includes an environmental perception subunit and a parameter adaptive adjustment subunit; wherein, The environmental perception subunit is used to collect ambient light intensity, image signal-to-noise ratio, and scene background complexity parameters in real time. The parameter adaptive adjustment subunit is used to generate control commands based on environmental perception parameters and target information output by the intelligent vision processing module, and adaptively adjust the laser output power, repetition rate, divergence angle, and pulse width of the active illumination module, as well as the detector exposure time, gain, frame rate, and imaging objective focal length of the coaxial imaging acquisition module.

[0013] Preferably, it further includes a servo tracking module; wherein, The servo tracking module includes a two-dimensional high-precision servo turntable, a turntable drive circuit, and an encoder; wherein... The servo tracking module is connected to the intelligent vision processing module and is used to receive the azimuth and pitch angle deviation information of the target, drive the two-dimensional servo turntable to rotate in real time, and make the system optical axis continuously aligned with the cat's eye target. The azimuth adjustment range of the two-dimensional servo turntable is ±170°, the pitch adjustment range is -45° to +85°, and the positioning accuracy is ≤0.05°.

[0014] Preferably, the display, control, and storage module communicates with the intelligent vision processing module and the adaptive control module via gigabit Ethernet; wherein, The touch screen of the display and storage module displays the active imaging image, the bounding box annotation of the cat eye target, the target category, confidence level, azimuth angle, distance, and motion trajectory information in real time; The display, control and storage module provides a human-machine interface that supports switching detection modes, setting parameters, adjusting focus, and inputting target locking commands. The display, control and storage module has a built-in storage unit that supports real-time storage, playback and export of raw image data, target detection data and system operation logs.

[0015] A second aspect of this application provides a method for an active imaging system for detecting cat-eye targets based on intelligent vision, the method comprising: S1: The system is powered on and initialized, completing self-tests of each module, parameter calibration, and loading of the deep learning network model. It also receives the user-set detection mode and initial parameters. S2: The adaptive control module sets the initial laser parameters of the active illumination module and the initial imaging parameters of the coaxial imaging acquisition module according to the current ambient light intensity. S3: The active illumination module emits pulsed laser light, which illuminates the target scene through the coaxial optical path of the coaxial imaging acquisition module. The back reflection echo of the target scene returns along the original optical path, is received by the coaxial imaging acquisition module, and the image is acquired and transmitted to the intelligent vision processing module. S4: The intelligent vision processing module performs preprocessing, candidate region extraction, and deep learning classification and recognition on the image data in sequence, distinguishing between real cat eye targets and false interference, and outputting the target's category, confidence level, and coordinate information; it also completes the target's positioning and ranging and multi-frame tracking simultaneously, and outputs the target's three-dimensional position information and motion trajectory. S5: The adaptive control module adjusts the laser parameters of the active lighting module and the imaging parameters of the coaxial imaging acquisition module in real time based on the current scene perception data and target information to form a closed-loop control. S6: The display, control and storage module displays the imaging screen and target information in real time and stores relevant data synchronously; if the tracking mode is enabled, the servo tracking module drives the turntable to continuously lock onto and track the target. S7: Repeat steps S3-S6 until a stop probe command is received.

[0016] Preferably, in step S5, the adjustment strategy of the adaptive control module includes: When detecting distant targets, increase the laser output power, decrease the laser divergence angle, extend the detector exposure time, and reduce the frame rate; When in a strong light complex background, reduce laser power, increase detector frame rate, and enable multi-frame background difference suppression; When tracking dynamic targets, increase the laser repetition rate and detector frame rate.

[0017] Therefore, this application has the following beneficial effects: (1) In this application, the candidate region extraction unit integrates adaptive threshold segmentation and connected component analysis to quickly screen suspected candidate regions; the deep learning feature extraction and classification recognition unit has a built-in lightweight CNN classification recognition network to perform deep feature extraction on the candidate region image slices. It completely breaks through the limitations of traditional algorithms that rely solely on brightness and size as single features for recognition, and comprehensively extracts multi-dimensional core features such as target roundness, edge gradient, diffraction texture, and spatial spectrum, effectively identifying various false interference sources such as specular reflection, metallic reflection, glass curtain walls, and vegetation highlights. At the same time, through diversified data enhancement methods such as rotation, scaling, noise addition, and brightness adjustment, combined with large-scale dataset training, the false alarm rate is significantly reduced. In addition, the system can accurately distinguish various types of cat-eye targets such as camera lenses, telescopes, sniper scopes, UAV optoelectronic pods, and vehicle-mounted optical equipment, providing detailed and reliable target attribute basis for subsequent threat level assessment and precise handling, and realizing refined classification and recognition of cat-eye targets; (2) This application adopts a strictly coaxial optical path design for the active illumination module and the coaxial imaging acquisition module. Specifically, by using a polarizing beam splitter and a quarter-wave plate, the illumination optical path and the imaging optical path are completely overlapped, eliminating the field-of-view difference between the illumination optical path and the imaging optical path. This achieves full-range precise matching between the laser illumination area and the imaging field of view, eliminating the detection blind zone during long-distance detection. At the same time, the coaxial design maximizes the collection of back-reflected echoes from the cat's-eye effect, significantly reducing echo loss during long-distance detection, significantly improving the echo energy utilization rate and the system signal-to-noise ratio, and effectively extending the maximum detection distance and detection sensitivity of the system.

[0018] (3) This application introduces an adaptive control module to construct a full-parameter adaptive closed-loop control link. This module collects environmental perception parameters such as ambient light intensity, image signal-to-noise ratio, and scene background complexity in real time, and combines them with information such as target distance and motion state output by the intelligent vision processing module to dynamically adjust the laser output power, repetition rate, divergence angle, and pulse width of the active illumination module, as well as the detector exposure time, gain, frame rate, and imaging objective focal length of the coaxial imaging acquisition module; Specific adjustment strategies include: when detecting distant targets, automatically increasing laser power and decreasing the divergence angle, while extending exposure time and reducing frame rate; when in strong light and complex backgrounds, automatically reducing laser power, increasing frame rate, and enabling multi-frame background difference suppression; and when tracking dynamic targets, automatically increasing laser repetition rate and detector frame rate. This closed-loop control chain enables the system to maintain high signal-to-noise ratio imaging in complex environments such as strong light, backlight, rain, fog, and haze, significantly improving detection robustness in all weather and complex environments. (4) The intelligent vision processing module of this application has a built-in target positioning and ranging unit. Based on the TOF principle, it accurately calculates the straight-line distance of the target and obtains the azimuth and pitch angles of the target by combining imaging intrinsic parameters, and outputs the three-dimensional position information of the target. At the same time, the multi-frame correlation tracking unit adopts the improved DeepSORT multi-target tracking algorithm combined with Kalman filtering to perform feature matching and trajectory correlation on the cat-eye targets in continuous frames, realizes continuous and stable tracking of static and dynamic cat-eye targets, and outputs the motion trajectory of the target. When the system is equipped with a servo tracking module, it can drive the two-dimensional servo turntable to rotate in real time according to the target angle deviation information, so that the optical axis of the system is continuously aligned with the cat-eye target, realizes stable locking of dynamic targets such as moving UAVs and mobile concealed observation equipment, and provides continuous and accurate three-dimensional position data and motion trajectory data for subsequent threat disposal and fire guidance. (5) This application adopts a modular, loosely coupled hardware and software design, with clear interfaces and dependencies between functional modules. The intelligent vision processing module adopts an FPGA+ARM heterogeneous computing architecture or an edge GPU computing module to complete all computing tasks in real time on a low-power, miniaturized platform. The system can flexibly tailor functional modules according to different application scenarios. For example, the servo tracking module can be omitted in handheld portable scenarios, the complete function can be integrated in fixed-point security, and customized configurations can be made for vehicle-mounted, airborne, and shipborne platforms according to space and power consumption constraints. The display, control, and storage module provides an intuitive human-machine interface and supports real-time storage, playback, and export of raw image data, target detection data, and system operation logs. This design gives the system strong versatility and engineering application value, and it can be widely adapted to various platforms such as handheld portable, fixed-point security, vehicle-mounted, airborne, and shipborne.

[0019] This solves the problems of high false alarm rate, limited detection range, poor environmental adaptability, and inability to achieve fine target classification and continuous tracking in existing technologies.

[0020] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0021] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a schematic diagram of the structure of an active imaging system for cat eye target detection based on intelligent vision, according to an embodiment of this application. Figure 2 This is a schematic diagram of the structure of an active lighting module according to an embodiment of this application; Figure 3 This is a schematic diagram of the optical path structure of a coaxial imaging acquisition module and an active illumination module according to an embodiment of this application; Figure 4 This is a schematic diagram of the structure of a servo tracking module according to an embodiment of this application; Figure 5 This is a flowchart of an active imaging method for cat eye target detection based on intelligent vision, according to an embodiment of this application. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] The following description, with reference to the accompanying drawings, illustrates an active imaging system and method for detecting cat-eye targets based on intelligent vision, according to an embodiment of this application. Addressing the high false alarm rate issue mentioned in the background section, this application provides an active imaging system for detecting cat-eye targets based on intelligent vision. In this system, by constructing a two-level detection architecture, it overcomes the limitations of traditional algorithms that rely solely on single features such as brightness and size. It comprehensively extracts multi-dimensional features such as target roundness, edge gradient, diffraction texture, and spatial spectrum. Combined with data augmentation and large-sample training, the false alarm rate is reduced to below 0.1%, and the recognition accuracy is increased to over 99%. Furthermore, it achieves refined classification and recognition of cat-eye targets such as camera lenses, sniper scopes, and UAV electro-optical pods. Employing a strictly coaxial polarization beam splitter design, using a polarization beam splitter prism and a quarter-wave plate, it eliminates the field-of-view difference between the illumination and imaging beam paths, achieving precise matching of the laser illumination area and the imaging field of view across the entire range, significantly improving echo energy. The system effectively extends the maximum detection range and sensitivity by improving utilization and system signal-to-noise ratio. An adaptive control module is introduced to construct a full-parameter closed-loop control link, which collects parameters such as ambient light intensity, signal-to-noise ratio, and background complexity in real time. Combined with target distance and motion state, the system dynamically adjusts laser power, repetition rate, divergence angle, detector exposure, gain, frame rate, and focal length, ensuring high signal-to-noise ratio imaging even in complex environments such as strong light, backlight, rain, fog, and haze. It integrates TOF high-precision positioning and ranging with an improved DeepSORT multi-target tracking algorithm, outputting the target's three-dimensional position information and motion trajectory. This, combined with a servo turntable, enables stable locking and continuous tracking of dynamic targets. A modular, low-coupling design is adopted, based on an FPGA+ARM or edge GPU heterogeneous computing architecture, allowing for flexible adaptation to various platforms such as handheld portable devices, fixed security systems, vehicle-mounted systems, airborne systems, and shipborne systems. Therefore, this application effectively solves the problems of high false alarm rate, limited detection range, poor environmental adaptability, and inability to achieve fine-grained classification and continuous tracking in existing technologies.

[0024] Figure 1 This is a schematic diagram of the structure of an active imaging system for cat-eye target detection based on intelligent vision, provided in an embodiment of this application.

[0025] This application provides an active imaging system for cat eye target detection based on intelligent vision. The system 10 includes: an active illumination module 100, a coaxial imaging acquisition module 200, an intelligent vision processing module 300, an adaptive control module 400, and a display, control, and storage module 500.

[0026] Among them, the active illumination module 100 is used to output pulsed laser with adaptively adjustable parameters to actively illuminate the target scene and stimulate the cat's eye effect back reflection echo of the optical device in the target scene; The coaxial imaging acquisition module 200 and the active illumination module 100 adopt a strictly coaxial optical path design for receiving and transmitting back the back reflection echo of the target scene, completing photoelectric conversion and image acquisition, and transmitting the acquired image data to the intelligent vision processing module 300. The intelligent vision processing module 300 is connected to the coaxial imaging acquisition module 200, the adaptive control module 400, and the display, control and storage module 500 respectively. It is used to perform intelligent vision processing on the received image data, complete the identification, positioning and tracking of the cat eye target, and output the target information and scene perception data. The adaptive control module 400 is connected to the active lighting module 100, the coaxial imaging acquisition module 200, and the intelligent vision processing module 300 respectively. It is used to adaptively adjust the laser output parameters of the active lighting module 100 and the imaging acquisition parameters of the coaxial imaging acquisition module 200 according to the scene perception data and target information, and construct a closed-loop control link. The display, control and storage module 500 is used to realize human-computer interaction, display of real-time imaging images and target information, and storage of raw images, target data and system logs.

[0027] It is understood that in this embodiment, the active lighting module, as the front-end excitation unit, uses adjustable parameter pulsed laser to achieve active scene illumination and stably excite the cat's eye back reflection characteristic signal of the optical device; the coaxial imaging acquisition module relies on a strictly coaxial optical path structure for transmission and reception to accurately match the laser illumination field of view, acquire back reflection echoes without blind spots and complete photoelectric conversion, and transmit high-quality image data to the intelligent vision processing module in real time; the intelligent vision processing module, as the core processing unit of the system, relies on the received image data to complete target screening, fine classification, position calculation and continuous tracking, and outputs target status information and scene environment parameters in real time. On the one hand, it pushes the data upward to the display, control and storage module for visualization and data retention, and on the other hand, it inputs the data downward to the adaptive control module as the basis for regulation; the adaptive control module relies on the multi-dimensional information such as scene complexity, signal-to-noise ratio, target distance and motion state perceived by the front end to correct the working parameters such as laser power, repetition rate and divergence angle of the active lighting module and the imaging parameters such as exposure, gain and frame rate of the coaxial imaging acquisition module in real time, so that the system is always in the optimal working state; the display, control and storage module realizes human-computer interaction operation, real-time image presentation and full-process data storage and traceability.

[0028] As one embodiment of this application, such as Figure 2 As shown, the active illumination module 100 includes a multi-band laser source, a beam shaping unit, an acousto-optic modulator, and a laser driving circuit; wherein, The multi-band laser source uses a near-infrared pulsed laser, with an output wavelength covering at least one of 808nm, 940nm, and 1064nm, to achieve covert active illumination. The beam shaping unit includes a collimating lens group and an adjustable beam expander, which are used to collimate and expand the laser beam to achieve adaptive adjustment of the laser divergence angle in the range of 0.1mrad to 5mrad. The acousto-optic modulator is used to modulate laser pulses and output nanosecond-level narrow pulse lasers, with an adjustable pulse repetition rate range of 10Hz to 10kHz. The laser drive circuit is connected to the adaptive control module to receive control commands and adjust the laser's output power, repetition rate, pulse width, and divergence angle.

[0029] It is understood that the active illumination module in this application adopts a subdivided unit architecture design. Through the cooperation and coordinated operation of multi-band laser light sources, beam shaping units, acousto-optic modulators, and laser driving circuits, a pulsed laser emission system with real-time adjustable parameters, selectable bands, and controllable beam quality is constructed. The multi-band laser light source uses a near-infrared pulsed laser, preferably a 940nm eye-safe pulsed fiber laser, with a maximum output power of up to 20W. This avoids the defects of easy exposure in the visible light band, ensuring eye safety while achieving all-weather concealed active illumination of the scene and efficiently stimulating the cat-eye effect back reflection echo of optical devices in the scene. The beam shaping unit is equipped with a collimating lens group and an electrically adjustable beam expander, which can accurately collimate and continuously expand and shape the laser beam. It can adaptively adjust the laser divergence angle within the range of 0.1mrad to 3mrad, accurately matching the imaging field of view at different distances, and avoiding detection blind spots caused by beam energy divergence waste or field of view matching deviation. The acousto-optic modulator enables nanosecond-level narrow-pulse laser output, controlling the pulse width to the 10ns level. The pulse repetition rate can be flexibly adjusted within a wide range of 10Hz to 5kHz, effectively compressing the laser pulse time domain width and improving the recognition of echo signals and the accuracy of detection and ranging. The laser drive circuit adopts a high-precision constant current drive chip and establishes a communication connection with the adaptive control module through an RS485 bus. It can receive closed-loop control commands in real time and dynamically adjust the laser output power, repetition rate, pulse width, and divergence angle, enabling the active illumination module to adaptively match the optimal operating parameters according to the ambient light intensity, target distance, and background complexity. This effectively improves the long-range detection capability, the ability to resist environmental interference, and the overall detection signal-to-noise ratio of the system.

[0030] As one embodiment of this application, such as Figure 3 As shown, the coaxial imaging acquisition module 200 includes a polarizing beam splitter, a quarter-wave plate, an imaging objective lens group, a narrow-band filter, a planar photodetector, and an image acquisition circuit.

[0031] The polarizing beam splitter, in conjunction with the quarter-wave plate, forms a coaxial optical path for transmitting and receiving: the laser output from the active illumination module is reflected by the polarizing beam splitter and then emitted to the target scene through the imaging objective lens group; the back reflection echo from the target scene returns along the original optical path, and after passing through the imaging objective lens group, the polarizing beam splitter, and the quarter-wave plate, it is transmitted and enters the narrowband filter. The center wavelength of the narrowband filter is matched with the laser wavelength of the active illumination module, and the half-width at half-maximum (WHM) is ≤10nm, which is used to filter out stray light from the ambient background. The imaging objective lens group uses a telephoto zoom objective lens with an adjustable focal length range of 50mm to 500mm and an adjustable field of view range of 0.1° to 5°. The area array photodetector uses a near-infrared enhanced CMOS / CCD / InGaAs detector, supports global exposure, and has an adjustable frame rate range of 1fps~60fps; The image acquisition circuit is used to perform analog-to-digital conversion and preprocessing of images, and transmits the acquired image data to the intelligent vision processing module.

[0032] It is understood that the embodiments of this application utilize the polarization matching characteristics of the polarization beam splitter and the quarter-wave plate to make the linearly polarized laser output by the active illumination module reflected by the S-plane of the polarization beam splitter and then converted into circularly polarized light by the quarter-wave plate before being emitted outward; the back-reflected echo of the cat's eye target is folded back along the original optical path and converted into linearly polarized light in the P direction by the quarter-wave plate again, which can pass through the polarization beam splitter without loss and enter the subsequent optical path. This not only achieves complete coaxial alignment between the illumination optical path and the imaging optical path, eliminating the field of view deviation and long-distance detection blind zone in the traditional split optical path, but also improves the return light utilization rate by more than 80%, while effectively suppressing non-target stray reflection interference.

[0033] The narrowband filter features a 940nm center wavelength design that matches the laser's operating wavelength, with a full width at half maximum (FWHM) of 8nm and an optical density (OD) ≥ 6. This allows for precise transmission of the target's echo laser while significantly filtering out stray light from ambient backgrounds such as natural light and artificial light, resulting in a marked improvement in the signal-to-noise ratio and target contrast of the image. The imaging objective lens employs a motorized telephoto zoom mechanism, with a focal length continuously adjustable from 100mm to 500mm, a field of view covering 0.1° to 3°, and support for autofocus. It can flexibly switch between telephoto and wide-angle fields of view depending on the target distance, accommodating both detailed detection of small targets at long distances and large-scale scene monitoring at close range.

[0034] The area-array photodetector uses a near-infrared enhanced global exposure CMOS detector with a pixel size of 5μm, a resolution of 1920×1080, and an adjustable frame rate range of 1fps to 30fps. It boasts a quantum efficiency of no less than 65% in the 940nm operating band, offering advantages such as high near-infrared response, no global ghosting, and a large dynamic range. It is suitable for stable acquisition of weak signals from low-light and distant cat's-eye echoes. The image acquisition circuit uses an FPGA as its core architecture, performing image analog-to-digital conversion, real-time caching, and front-end preprocessing. Image data is then transmitted at high speed to the intelligent vision processing module via gigabit Ethernet, ensuring real-time, complete, and low-latency image transmission. This provides high-quality raw imaging data support for subsequent target recognition, positioning, and adaptive closed-loop control.

[0035] As one embodiment of this application, the intelligent vision processing module 300 adopts an FPGA+ARM heterogeneous computing architecture or an edge GPU computing module, including an image preprocessing unit, a candidate region extraction unit, a deep learning feature extraction and classification recognition unit, a target localization and ranging unit, and a multi-frame correlation tracking unit; wherein... The image preprocessing unit is used to sequentially perform blind pixel correction, non-uniformity correction, adaptive denoising, contrast enhancement, and background difference suppression on the received raw image. The candidate region extraction unit is used to integrate adaptive threshold segmentation connected component analysis with a lightweight target detection network to quickly filter out candidate regions of suspected cat-eye targets and generate candidate target boxes. The deep learning feature extraction and classification unit has a built-in lightweight CNN classification network, which is used to extract features from image slices of candidate regions, distinguish between real cat eye targets and false interference sources, and output the category and corresponding confidence of the cat eye target. The target localization and ranging unit is used to calculate the pixel coordinates of the target in the image based on the identified cat eye target, obtain the azimuth and pitch angles of the target by combining the imaging intra-participant calibration parameters, and calculate the straight-line distance of the target based on the laser time-of-flight (TOF) principle, and output the three-dimensional position information of the target. The multi-frame correlation tracking unit uses an improved DeepSORT multi-target tracking algorithm, combined with Kalman filtering, to perform feature matching and trajectory correlation on cat-eye targets in consecutive frames, and outputs the motion trajectory of the targets.

[0036] It is understood that the intelligent vision processing module in this application embodiment adopts an FPGA+ARM heterogeneous computing architecture or an edge GPU computing module. It is divided into an image preprocessing unit, a candidate region extraction unit, a deep learning feature extraction and classification recognition unit, a target localization and ranging unit, and a multi-frame correlation tracking unit. Each unit works in sequence according to the pipeline timing to complete the entire process of intelligent vision processing from original image input to target recognition, 3D localization, and continuous trajectory tracking.

[0037] The image preprocessing unit first performs blind pixel correction, non-uniformity correction, adaptive denoising, contrast enhancement, and background subtraction suppression on the received raw image in sequence. This effectively eliminates detector imaging defects, environmental noise, and complex background interference, improving the overall image quality and target feature saliency, and providing clean and reliable raw image data for subsequent target detection and recognition. The candidate region extraction unit integrates adaptive threshold segmentation, connected component analysis, and a lightweight target detection network. It can quickly filter out suspected cat-eye target candidate regions from the entire scene image and generate candidate target boxes, significantly reducing the computational scope for subsequent fine recognition, and balancing detection real-time performance and candidate target recall rate.

[0038] The deep learning feature extraction and classification unit incorporates a lightweight CNN classification network, trained on a self-built cat-eye target dataset containing hundreds of thousands of samples. The samples cover various detection distances from 50m to 3000m, wide dynamic range lighting from 0 to 10000 lux, and various complex indoor and outdoor backgrounds. It also includes a large number of false interference samples such as specular reflections, metallic reflections, and vegetation highlights. Data augmentation techniques such as rotation, scaling, noise addition, brightness and contrast transformations are used to expand sample diversity. The model is optimized using a cross-entropy loss function. The trained network can accurately extract deep features from candidate region image slices, effectively distinguishing real cat-eye targets from various false interference sources, and outputting the target category and confidence level. The recognition accuracy reaches over 99.2%, with a false alarm rate controlled below 0.08%.

[0039] After identifying the real cat-eye target, the target positioning and ranging unit accurately calculates the centroid pixel coordinates of the target image. Combined with the imaging objective's internal participation in the system calibration extrinsic parameters, it calculates the azimuth and elevation angles of the target relative to the system's optical axis. Simultaneously, based on the laser time-of-flight (TOF) principle, it accurately and synchronously records the laser emission and echo reception times, calculates the target's straight-line distance through the time of flight, and finally outputs complete three-dimensional position information. The ranging accuracy is controlled within 1%, meeting the requirements for high-precision target position calculation.

[0040] The multi-frame correlation tracking unit uses an improved DeepSORT multi-target tracking algorithm combined with Kalman filtering to perform appearance feature matching and trajectory correlation on cat-eye targets in consecutive frames. It can simultaneously and stably track multiple static and dynamic cat-eye targets and output complete motion trajectories. It can also effectively filter out occasional false alarm interference in single frames, further suppressing false alarms from a temporal perspective, and significantly improving the continuity, stability and reliability of the detection and recognition results of the entire system.

[0041] In this embodiment, the training dataset of the lightweight CNN classification and recognition network includes real cat eye target samples and various fake interference samples with different distances, different lighting, and different backgrounds. The dataset is expanded by data augmentation methods such as rotation, scaling, noise addition, and brightness adjustment. The lightweight CNN classification network achieves an accuracy of ≥99% in recognizing cat-eye targets, with a false alarm rate of ≤0.1%. The lightweight CNN classification network can identify cat-eye target categories including at least one of camera lenses, telescopes, sniper scopes, drone electro-optical pods, and vehicle-mounted optical equipment.

[0042] It should be noted that the lightweight CNN network structure used in this application is not an undisclosed, arbitrary design, but rather a fine-tuning and adaptation based on open-source lightweight backbone networks such as MobileNetV2 or ShuffleNetV2, specifically tailored to the visual characteristics of cat-eye targets: "small size, high brightness, and specific diffraction texture." Those skilled in the art can select any of the aforementioned lightweight networks as the feature extraction backbone based on the computational constraints of the actual deployment platform. Guided by the training dataset disclosed in this application (distances from 50m to 3000m, illumination from 0 to 10000 lux, covering complex indoor and outdoor backgrounds and false interference samples, totaling hundreds of thousands of samples) and data augmentation methods (rotation, scaling, noise addition, brightness and contrast transformation), a classification model satisfying a recognition accuracy ≥99% and a false alarm rate ≤0.1% can be trained using the cross-entropy loss function. This application has disclosed the training data range, data augmentation methods, loss function type, and performance targets. Those skilled in the art can reproduce the training and deployment of this network model based on the above guidance without additional creative effort.

[0043] The DeepSORT algorithm is a publicly available and widely used multi-object tracking framework in the field of computer vision. The improvements in this application are mainly reflected in the following aspects: For the characteristics of cat-eye targets, such as echo flicker, brief occlusion, or sudden brightness changes in consecutive frames, the trajectory survival threshold and appearance feature update strategy in cascaded matching are adjusted. Simultaneously, a lightweight constraint on the prior distribution of the cat-eye target's motion velocity is introduced into the Kalman filter motion model. Specific parameter settings can be obtained by those skilled in the art through a limited number of field experiments or offline calibrations during the system initialization phase, depending on the detection scenario (such as fixed zone security, vehicle-mounted dynamic detection, UAV-borne tracking, etc.).

[0044] The TOF principle itself is a well-known technology in the field of lidar ranging. The ranging accuracy described in this application is controlled within 1%, which is achieved through the following engineering means: the acousto-optic modulator of the active illumination module provides nanosecond-level narrow pulse laser (typical pulse width of 5ns to 20ns), the image acquisition circuit in the coaxial imaging acquisition module synchronously records the laser emission time and the echo arrival time, and the intelligent vision processing module performs high-precision time delay calculation.

[0045] As one embodiment of this application, the adaptive control module 400 includes an environment perception subunit and a parameter adaptive adjustment subunit; wherein... The environmental perception subunit is used to collect ambient light intensity, image signal-to-noise ratio, and scene background complexity parameters in real time; The parameter adaptive adjustment subunit is used to generate control commands based on environmental perception parameters and target information output by the intelligent vision processing module, and adaptively adjust the laser output power, repetition rate, divergence angle, and pulse width of the active illumination module, as well as the detector exposure time, gain, frame rate, and imaging objective focal length of the coaxial imaging acquisition module.

[0046] It is understood that the embodiments of this application use an STM32H7 series MCU as the main control core. The environmental perception subunit collects ambient light intensity information in real time through a light sensor and relies on the intelligent vision processing module to obtain multi-dimensional perception parameters such as image signal-to-noise ratio, scene background complexity, number of targets, target distance, and target movement speed in real time, providing a comprehensive and realistic basis for subsequent adaptive control. The parameter adaptive adjustment subunit intelligently generates precise control commands based on the collected environmental perception parameters and target status information, and can simultaneously adjust multiple key parameters of the front-end equipment. It can adaptively control the laser output power, repetition rate, divergence angle, and pulse width of the active illumination module, and can also dynamically configure the detector exposure time, gain, frame rate, and imaging lens focal length of the coaxial imaging acquisition module.

[0047] For example, when detecting targets at a distance of over 1000m, the laser output power is automatically increased to 20W, the laser divergence angle is adjusted to 0.1mrad, and the detector exposure time is extended while the frame rate is reduced to 5fps, effectively enhancing the strength of weak echo signals at long distances and improving the imaging signal-to-noise ratio. When in a complex background with strong light and high reflectivity, the laser power is automatically reduced to 5W, the detector frame rate is simultaneously increased to 30fps, and a multi-frame background difference suppression algorithm is enabled to effectively reduce interference from background stray light and high reflectivity. When tracking dynamic targets with a speed exceeding 5m / s, the laser repetition rate is automatically increased to 5kHz and the detector frame rate is increased to 30fps to ensure the real-time and continuous tracking of the target. At the same time, the imaging objective lens focal length and focus position can be adaptively adjusted according to changes in the distance of the target to ensure that the target image is always clear and stable.

[0048] As one embodiment of this application, such as Figure 4 As shown, it also includes a servo tracking module; wherein, The servo tracking module includes a two-dimensional high-precision servo turntable, a turntable drive circuit, and an encoder; among which, The servo tracking module is connected to the intelligent vision processing module to receive the target's azimuth and pitch deviation information, drive the two-dimensional servo turntable to rotate in real time, and keep the system's optical axis aligned with the cat's eye target. The azimuth adjustment range of the two-dimensional servo turntable is ±170°, the pitch adjustment range is -45° to +85°, and the positioning accuracy is ≤0.05°.

[0049] It is understood that the embodiments of this application consist of a two-dimensional high-precision servo turntable, a turntable drive circuit, and an absolute encoder. The whole system establishes a data interaction connection with the intelligent vision processing module to form a complete target closed-loop tracking and control system. For example, the two-dimensional servo turntable has a large angle adjustment range, with an azimuth adjustment range of ±170°, a pitch adjustment range covering -45° to +85°, an angle resolution of up to 0.005°, and an overall positioning accuracy of ≤0.05°. It has an ultra-large field of view and ultra-high angle control accuracy, which can meet the needs of large-scale, all-airspace target search and tracking.

[0050] During operation, the turntable drive circuit receives target azimuth and pitch angle deviation information from the intelligent vision processing module in real time. Combined with angle feedback data from the high-precision encoder, a closed-loop precise drive is achieved through a PID control algorithm. This controls the two-dimensional servo turntable to rotate smoothly and continuously, dynamically correcting the system's optical axis pointing deviation. This ensures the system's detection optical axis remains consistently aligned with the center of the target, enabling automatic locking, continuous following, and stable tracking of both static fixed targets and high-speed dynamic moving targets. This module effectively solves the problems of traditional detection systems having fixed fields of view, being unable to follow dynamic targets, and being prone to missing distant targets. It significantly improves the system's continuous detection stability and tracking reliability against dynamic threat targets such as UAV optoelectronic devices and mobile concealed observation optical devices, providing continuous, stable, and high-precision angle pointing data support for subsequent accurate threat assessment, trajectory recording, and fire guidance.

[0051] As one embodiment of this application, the display control and storage module 500 communicates with the intelligent vision processing module and the adaptive control module via Gigabit Ethernet; wherein... The touch screen of the display, control and storage module 500 displays the active imaging image, the bounding box annotation of the cat eye target, the target category, confidence level, azimuth angle, distance and motion trajectory information in real time; The display, control and storage module 500 provides a human-machine interface that supports detection mode switching, parameter setting, focus adjustment, and target locking command input. The display, control and storage module 500 has a built-in storage unit that supports real-time storage, playback and export of raw image data, target detection data and system operation logs.

[0052] It is understood that this application embodiment establishes high-speed data communication links with the intelligent vision processing module and the adaptive control module via gigabit Ethernet to ensure the real-time performance and stability of data transmission. It consists of a 10.1-inch industrial-grade touchscreen display, an industrial control host computer, and a 4TB solid-state storage hard drive, integrating visualization, human-machine interaction, data storage, and data traceability. The industrial-grade touchscreen display can load and display the system's active imaging images in real time, accurately marking detected targets with red frames, and simultaneously and intuitively presenting comprehensive detection information such as target category, recognition confidence level, azimuth angle, pitch angle, straight-line distance, and motion trajectory, facilitating real-time monitoring of target status and detection scene conditions by personnel. Simultaneously, the module is equipped with a comprehensive human-machine interface, supporting switching between handheld and fixed detection modes, and enabling manual configuration of laser parameters and imaging parameters, adjustment of imaging focus, manual target locking, system calibration, and other operation commands. The human-machine operation is convenient and flexible, adapting to different operational needs. In addition, the module's built-in 4TB solid-state storage hard drive can complete the real-time storage of raw image data, full target detection data, and system operation logs. It has ample storage space and strong storage stability. It also supports local data playback, retrieval, and USB flash drive export functions, enabling full traceability and review of detection and operation data. This provides complete and reliable data support for system debugging, data analysis, job archiving, and subsequent technology iteration and scenario optimization.

[0053] This application provides an active imaging system for cat-eye target detection based on intelligent vision. It comprehensively extracts multi-dimensional features such as target circularity, edge gradient, diffraction texture, and spatial spectrum. Combined with data augmentation and large-sample training, it reduces the false alarm rate to below 0.1% and improves the recognition accuracy to over 99%. It also enables refined classification and recognition of cat-eye targets such as camera lenses, sniper scopes, and UAV electro-optical pods. The system adopts a polarization beam splitter optical path design with strict coaxial transmission and reception. By using a polarization beam splitter prism and a quarter-wave plate, it eliminates the field-of-view difference between the illumination optical path and the imaging optical path, achieving full-range accurate matching between the laser illumination area and the imaging field of view. This significantly improves the echo energy utilization rate and the system signal-to-noise ratio, and effectively extends the maximum detection distance and detection sensitivity. An adaptive control module is introduced to construct a full-parameter closed-loop control link, which collects parameters such as ambient light intensity, signal-to-noise ratio, and background complexity in real time. Combined with target distance and motion state, the system dynamically adjusts laser power, repetition rate, divergence angle, detector exposure, gain, frame rate, and focal length, enabling the system to maintain high signal-to-noise ratio imaging even in complex environments such as strong light, backlight, rain, fog, and haze. It integrates TOF high-precision positioning and ranging with an improved DeepSORT multi-target tracking algorithm, outputting the target's three-dimensional position information and motion trajectory. This, combined with a servo turntable, achieves stable locking and continuous tracking of dynamic targets. A modular, loosely coupled design is adopted, based on an FPGA+ARM or edge GPU heterogeneous computing architecture, allowing for flexible adaptation to various platforms such as handheld portable devices, fixed security systems, vehicle-mounted systems, airborne systems, and shipborne systems. Therefore, this application effectively solves the problems of high false alarm rate, limited detection range, poor environmental adaptability, and inability to achieve fine-grained classification and continuous tracking in existing technologies.

[0054] The following detailed description, using specific embodiments, of the active imaging system for detecting cat eyes based on intelligent vision proposed in this application. This embodiment is specifically a fixed security detection system capable of active detection, accurate identification, 3D positioning, and continuous tracking of optical cat eye targets in complex outdoor scenes. Specific implementation details are as follows: The active illumination module comprises a multi-band laser source, a beam shaping unit, an acousto-optic modulator, and a laser drive circuit, providing the system with adjustable parameter concealed active laser illumination. The multi-band laser source uses a 940nm pulsed fiber laser in the human eye-safe band, with a maximum output power of 20W. This avoids the problem of visible light illumination easily exposing detection equipment, achieving all-weather concealed active illumination and efficiently stimulating the cat-eye effect back reflection echo from various optical devices within the scene. The beam shaping unit consists of a collimating lens group and a motorized adjustable beam expander, enabling precise collimation and continuous beam expansion control of the output laser beam. This allows the laser divergence angle to be continuously and adaptively adjusted within the range of 0.1mrad to 3mrad, precisely matching the imaging field of view according to different detection distances, avoiding beam energy waste and detection blind spots caused by field-of-view mismatch. The acousto-optic modulator employs high-performance nanosecond-level acousto-optic modulators, enabling 10ns-level narrow-pulse laser output. The pulse repetition rate is adjustable from 10Hz to 5kHz, and this narrow pulse characteristic effectively improves laser ranging accuracy and echo signal recognition. The laser drive circuit uses a high-precision constant-current drive chip and establishes a stable communication connection with the adaptive control module via an RS485 bus. This allows for real-time reception of closed-loop control commands, dynamically adjusting the laser's output power, repetition rate, pulse width, and divergence angle to ensure the laser illumination parameters are adaptable to various complex detection scenarios.

[0055] The coaxial imaging acquisition module's optical path structure strictly adopts a coaxial transmit-receive design, mainly including a polarizing beam splitter (PBS), a quarter-wave plate, an imaging objective lens group, a narrowband filter, a planar photodetector, and an image acquisition circuit. This enables blind-zone-free, high-utilization echo signal acquisition and imaging. In this embodiment, the polarizing beam splitter and the quarter-wave plate work together to form a complete coaxial transmit-receive optical path. The linearly polarized laser output from the active illumination module is reflected by the S-plane of the polarizing beam splitter, passes through the quarter-wave plate, and is converted into circularly polarized light. After being collimated by the imaging objective lens group, it is emitted to the target scene. The back-reflected echo generated by the cat's eye target in the target scene is reflected back along the original optical path, passes through the quarter-wave plate again, and is converted into P-direction linearly polarized light. This light can be completely transmitted through the polarizing beam splitter, effectively avoiding interference from the illumination optical path to the imaging optical path. Compared with traditional split optical path designs, this coaxial structure can improve the backlight utilization rate by more than 80% while significantly suppressing ineffective stray light interference. The narrowband filter, with its center wavelength matched to the laser's operating wavelength of 940nm and a full width at half maximum (FWHM) of only 8nm, boasts an optical density (OD) ≥ 6. This allows for precise transmission of target laser echoes and deep filtering of ambient stray light, significantly improving the imaging signal-to-noise ratio. The imaging objective lens employs a motorized zoom telephoto lens with an adjustable focal length range of 100mm–500mm and an adjustable field of view range of 0.1°–3°. It supports autofocus and can adapt to the clear imaging requirements of targets at varying distances. The area array photodetector utilizes a near-infrared enhanced global exposure CMOS detector with a pixel size of 5μm, a resolution of 1920×1080, and an adjustable frame rate range of 1fps–30fps. It achieves a quantum efficiency ≥ 65% in the 940nm operating band and exhibits high sensitivity to weak cat's-eye echo signals. The image acquisition circuit, built around an FPGA, performs high-speed analog-to-digital conversion, data caching, and front-end preprocessing. High-quality image data is transmitted in real-time with low latency to the intelligent vision processing module via Gigabit Ethernet.

[0056] The intelligent vision processing module is the core data processing and target solving unit of this system. In this embodiment, it adopts the NVIDIA Jetson Xavier NX edge GPU computing module, with a computing power of up to 21 TOPS, which fully meets the requirements of parallel real-time operation of multiple algorithms. The module has built-in image preprocessing unit, candidate region extraction unit, deep learning feature extraction and classification recognition unit, target localization and ranging unit, and multi-frame correlation tracking unit, realizing intelligent processing of the entire process from image optimization, target selection, accurate recognition to localization and tracking. Among them, the image preprocessing unit can sequentially perform blind pixel correction, non-uniformity correction, adaptive median filtering denoising, multi-scale Retinex contrast enhancement, and inter-frame background difference suppression on the received raw image, effectively eliminating detector imaging defects, environmental noise, and static background clutter, significantly improving the feature contrast of the target area, and providing a high-quality image data source for subsequent target detection and recognition. The candidate region extraction unit adopts a two-level screening mechanism of "traditional algorithm + lightweight network". First, it uses adaptive threshold segmentation combined with eight-neighbor connected component analysis to quickly extract high-brightness suspected regions in the image to generate first-level candidate boxes. Then, it uses a lightweight YOLOv8n target detection network to complete the fast detection of the whole image and generate second-level candidate boxes. The two-level results are fused and duplicate redundant boxes are removed to accurately output the cat-eye target candidate region. This can reduce the computation of subsequent algorithms by more than 90% and greatly improve the real-time performance of the system.

[0057] The deep learning feature extraction and classification unit incorporates a lightweight CNN classification network. The network structure includes 5 convolutional layers, 3 pooling layers, and 2 fully connected layers. It can automatically extract multi-dimensional depth features specific to cat-eye targets, such as roundness, edge gradient, diffraction texture, spatial spectrum, and echo intensity distribution. It achieves accurate discrimination through a softmax classifier, effectively distinguishing real cat-eye targets from various false interference sources such as mirror reflection, metal reflection, glass curtain walls, and vegetation highlights. At the same time, it can accurately identify five major categories of cat-eye targets: pinhole cameras, SLR camera lenses, telescopes, sniper scopes, and drone electro-optical pods, and output the corresponding recognition confidence scores. This embodiment uses a self-built 100,000-level cat-eye target dataset to complete model training. The samples cover detection distances of 50m to 3000m, wide dynamic range lighting of 0 to 10000 lux, and various complex indoor and outdoor backgrounds. At the same time, data augmentation methods such as rotation, scaling, noise addition, brightness and contrast transformation are used to expand the diversity of samples. The cross-entropy loss function is used to optimize the model. After training, the network recognition accuracy is ≥99.2% and the false alarm rate is ≤0.08%, which has a very strong anti-interference recognition capability.

[0058] The target localization and ranging unit, based on the identified real cat-eye target, accurately calculates the centroid pixel coordinates of the target image. Combined with the imaging objective's internal parameters for system calibration, it calculates the target's azimuth and elevation angles relative to the system's optical axis. Simultaneously, based on the laser TOF time-of-flight principle, it synchronously records the laser emission and echo reception times, calculates the laser flight time Δt, and uses the formula R=c×Δt / 2 to calculate the target's straight-line distance. Finally, it outputs the target's complete three-dimensional position information, with an overall ranging accuracy ≤1%, meeting high-precision detection requirements. The multi-frame correlation tracking unit employs an improved DeepSORT multi-target tracking algorithm combined with Kalman filtering to perform appearance feature matching and trajectory correlation on consecutive frames of cat-eye targets. This enables continuous and stable tracking of multiple static and dynamic targets, outputting the target's motion trajectory in real time. It also effectively filters out occasional false alarm targets in a single frame, further reducing the system's false alarm rate from a temporal perspective and significantly improving the stability and continuity of detection results.

[0059] The adaptive control module uses an STM32H7 series MCU as its main control core and is interconnected with the active lighting module, coaxial imaging acquisition module, and intelligent vision processing module via a high-speed bus to construct a full-parameter adaptive closed-loop control link. The module includes an environmental perception subunit and a parameter adaptive adjustment subunit. The environmental perception subunit collects ambient light intensity parameters in real time through a light sensor and simultaneously receives multi-dimensional operating condition information from the intelligent vision processing module, such as image signal-to-noise ratio, scene background complexity, number of targets, target distance, and target movement speed, to comprehensively perceive and detect the environment and target status in real time. The parameter adaptive adjustment subunit intelligently generates differentiated control commands based on real-time sensing data, achieving adaptive optimization of parameters across all scenarios: When detecting targets at a distance of over 1000m, it automatically increases the laser output power to 20W and adjusts the divergence angle to 0.1mrad, while extending the detector exposure time and reducing the frame rate to 5fps, enhancing weak echo signals at long distances and improving the imaging signal-to-noise ratio; When facing complex background scenes with strong light and high reflectivity, it automatically reduces the laser power to 5W, increases the detector frame rate to 30fps, and activates the multi-frame background difference suppression algorithm to effectively suppress background stray light and high-light interference; When tracking dynamic targets with a speed exceeding 5m / s, it automatically increases the laser repetition rate to 5kHz and the detector frame rate to 30fps to ensure real-time target tracking; At the same time, it can dynamically adjust the focal length and focus position of the imaging objective lens according to the target distance to always maintain clear target imaging, fully adapting to complex all-weather detection scenarios.

[0060] The display, control, and storage modules establish high-speed communication links with the intelligent vision processing module and adaptive control module via gigabit Ethernet. The entire module comprises a 10.1-inch industrial-grade touchscreen display, an industrial control host computer, and a 4TB solid-state drive, integrating visual monitoring, human-machine interaction, and data storage and traceability functions. The touchscreen display shows the system's active imaging in real time, accurately marking identified targets with red frames, and simultaneously visually presenting key detection information such as target category, recognition confidence level, azimuth angle, elevation angle, straight-line distance, and motion trajectory, allowing operators to monitor the detection status in real time. It also features a comprehensive human-machine interface, supporting various operation commands such as handheld / fixed detection mode switching, manual configuration of laser and imaging parameters, objective lens focus adjustment, manual target locking, and system calibration, making it convenient and adaptable to various operational scenarios. The built-in 4TB high-capacity solid-state drive can store raw image data, target detection results data, and the entire system operation log in real time. It also supports local data playback and USB flash drive export, enabling full traceability and review of detection data, providing reliable data support for system debugging, work archiving, and technological iteration.

[0061] The servo tracking module mainly includes a two-dimensional high-precision servo turntable, a turntable drive circuit, and an absolute encoder, enabling high-precision automatic locking and continuous tracking of cat-eye targets. In this embodiment, the azimuth adjustment range of the two-dimensional servo turntable is ±170°, the pitch adjustment range is -30° to +60°, the angle resolution can reach 0.005°, and the positioning accuracy is ≤0.05°, covering a wide range of airspace detection needs with extremely high angle control accuracy. During operation, the turntable drive circuit receives the target azimuth and pitch deviation information output by the intelligent vision processing module in real time. Combined with the real-time angle feedback data from the absolute encoder, the PID closed-loop control algorithm drives the two-dimensional servo turntable to rotate smoothly and accurately, dynamically correcting the system's optical axis pointing deviation. This ensures that the system's detection optical axis is always aligned with the center of the cat-eye target, enabling stable locking and continuous tracking of static targets and high-speed dynamic moving targets. This effectively solves the problems of traditional fixed-field-of-view detection systems being prone to missing targets and unable to track dynamic targets, significantly improving the stability and reliability of long-distance dynamic cat-eye target detection.

[0062] Next, referring to the accompanying drawings, an active imaging method for cat eye target detection based on intelligent vision, according to an embodiment of this application, is described.

[0063] like Figure 5 As shown, this active imaging method for cat eye target detection based on intelligent vision includes the following steps: S1: The system is powered on and initialized, completing self-tests of each module, parameter calibration, and loading of the deep learning network model. It also receives the user-set detection mode and initial parameters. S2: The adaptive control module sets the initial laser parameters of the active illumination module and the initial imaging parameters of the coaxial imaging acquisition module according to the current ambient light intensity. S3: The active illumination module emits pulsed laser light, which illuminates the target scene through the coaxial optical path of the coaxial imaging acquisition module. The back reflection echo of the target scene returns along the original optical path, is received by the coaxial imaging acquisition module, and the image is acquired and transmitted to the intelligent vision processing module. S4: The intelligent vision processing module performs preprocessing, candidate region extraction, and deep learning classification and recognition on the image data in sequence, distinguishing between real cat eye targets and false interference, and outputting the target's category, confidence level, and coordinate information; it also completes the target's positioning and ranging and multi-frame tracking simultaneously, and outputs the target's three-dimensional position information and motion trajectory. S5: The adaptive control module adjusts the laser parameters of the active lighting module and the imaging parameters of the coaxial imaging acquisition module in real time based on the current scene perception data and target information to form a closed-loop control. S6: The display, control and storage module displays the imaging screen and target information in real time and stores relevant data synchronously; if the tracking mode is enabled, the servo tracking module drives the turntable to continuously lock onto and track the target. S7: Repeat steps S3-S6 until a stop probe command is received.

[0064] Understandably, this embodiment first powers on to complete the overall system initialization, performs hardware self-tests and system parameter calibrations on each functional module, and loads and deploys the deep learning network model. Simultaneously, it receives the detection mode and initial operating parameters configured by the operator, laying the foundation for stable detection in the future. After initialization, the adaptive control module matches optimal operating parameters based on real-time ambient light intensity information, adaptively configuring the initial laser parameters for the active illumination module and the initial imaging parameters for the coaxial imaging acquisition module, ensuring the system adapts to the current environmental baseline conditions. Subsequently, the active illumination module emits adjustable parameter pulsed laser light, which uniformly illuminates the target scene via the coaxial optical path of the coaxial imaging acquisition module, stimulating the cat's eye effect back reflection echo from the optical devices within the scene. The echo signal is then accurately received by the coaxial imaging acquisition module after refracting along the original optical path. After photoelectric conversion and image preprocessing, high-quality image data is transmitted in real-time to the intelligent vision processing module. The intelligent vision processing module performs image preprocessing, rapid screening of suspected target candidate regions, and accurate classification and recognition using deep learning on the received image data. It accurately distinguishes between genuine cat-eye targets and various environmental false interference sources, stably outputting target category, recognition confidence level, and pixel coordinate information. Simultaneously, relying on the TOF ranging principle and multi-frame correlation tracking algorithm, it synchronously completes the target's 3D position calculation and motion trajectory fitting, outputting high-precision positioning and ranging data and dynamic trajectory information. Throughout the system's detection process, the adaptive control module continuously collects scene environment perception data and target status information, iteratively optimizing core parameters such as laser output power, repetition rate, divergence angle, imaging exposure, gain, frame rate, and focal length in real time, forming a continuous closed-loop adaptive control mechanism to maintain the system in optimal detection state. Meanwhile, the display, control, and storage module refreshes and displays the active imaging screen and all target detection information in real time, enabling human-machine visual interaction, and storing and archiving original images, target data, and system logs in real time. When the system enters tracking mode, the servo tracking module drives the turntable to fine-tune its attitude in real time based on target angle deviation data, achieving continuous locking and stable tracking of the cat-eye target. The system continuously cycles through imaging acquisition, intelligent analysis, adaptive control, display and storage, and servo tracking until it receives an external stop detection command and terminates its operation. This significantly improves the system's detection accuracy, recognition stability, and all-weather adaptability, making it highly practical for engineering applications and adaptable to various scenarios.

[0065] As an embodiment of this application, in step S5, the adjustment strategy of the adaptive control module includes: When detecting distant targets, increase the laser output power, decrease the laser divergence angle, extend the detector exposure time, and reduce the frame rate; When in a strong light complex background, reduce laser power, increase detector frame rate, and enable multi-frame background difference suppression; When tracking dynamic targets, increase the laser repetition rate and detector frame rate.

[0066] It is understood that the embodiments of this application have designed differentiated and scene-adaptive parameter adaptive adjustment strategies for different detection distances, different environmental backgrounds, and different target motion states, so as to realize the intelligent dynamic optimal configuration of system working parameters and effectively solve the technical problems of fixed parameters, poor adaptability to working conditions, low signal-to-noise ratio in complex scenes, and easy loss of dynamic targets in traditional detection systems.

[0067] Specifically, when the system detects distant cat-eye targets, it increases laser output power and reduces laser divergence angle to concentrate laser energy and enhance the intensity of long-distance echo radiation. Simultaneously, it extends detector exposure time and reduces imaging frame rate, effectively accumulating weak echo signals and significantly improving the signal-to-noise ratio and detection probability of distant weak targets. This overcomes the problems of severe signal attenuation and difficulty in target identification at long distances. When the system is in complex background conditions with strong light and high reflectivity, actively reducing laser output power avoids overexposure and regional saturation caused by excessive laser intensity. It also increases the detector imaging frame rate and, in conjunction with a multi-frame background difference suppression strategy, effectively eliminates static highlights, stray reflections, and other background interference textures, dynamically filters environmental noise, and significantly reduces the false alarm rate in complex, high-light scenarios. When the system tracks high-speed dynamic cat-eye targets, by simultaneously increasing the laser pulse repetition rate and the detector imaging frame rate, the system's data sampling frequency and target refresh rate can be effectively improved, the single-frame imaging interval can be shortened, and the problems of imaging ghosting, position jumps and trajectory disconnection of high-speed moving targets can be reduced, thus ensuring the real-time performance, continuity and stability of dynamic target tracking.

[0068] According to the embodiments of this application, an active imaging method for cat-eye target detection based on intelligent vision is proposed. This method constructs a fully automatic cat-eye detection system with active detection, intelligent identification, adaptive optimization, continuous tracking, and full traceability through the coordinated operation of initialization calibration, adaptive parameter configuration, coaxial active illumination imaging, intelligent visual recognition and positioning, multi-condition dynamic closed-loop control, and servo-following tracking. This method overcomes the limitations of traditional fixed-parameter detection modes, relying on real-time environmental perception and target status feedback to achieve dynamic matching of laser illumination parameters and imaging acquisition parameters. It can simultaneously adapt to extreme detection scenarios such as distant weak targets, strong light complex backgrounds, and high-speed dynamic targets, significantly improving the accuracy of cat-eye target recognition, positioning and ranging accuracy, and dynamic tracking stability, effectively reducing the system's false alarm rate and missed detection rate.

[0069] The following detailed description of the active imaging method for cat-eye target detection based on intelligent vision proposed in this application will be provided through another specific embodiment. This embodiment corresponds to a handheld portable detection system version and is mainly suitable for lightweight and mobile operation scenarios such as individual soldier covert reconnaissance in the field, indoor anti-spy camera investigation, and close-range optical equipment investigation. The specific implementation details are as follows: After powering on, the device can automatically complete full-module hardware self-test, precise system parameter calibration, and rapid loading and solidification of lightweight deep learning models. It supports custom detection mode switching and initial parameter configuration. Based on real-time environmental perception results, it can adaptively match the optimal working parameters for laser illumination and imaging acquisition. Relying on a strictly coaxial optical path structure for transmission and reception, it achieves efficient active illumination and echo signal acquisition, preserving weak cat-eye echo characteristics to the greatest extent. Through a multi-level intelligent algorithm process of image preprocessing, rapid candidate region screening, accurate deep learning classification and recognition, 3D positioning and ranging, and multi-frame correlation tracking, it accurately eliminates various environmental false interferences and stably outputs target category, confidence level, 3D coordinates, and motion trajectory. Based on the scene and target status, it dynamically iterates and adjusts the system working parameters in real time. With human-computer interaction display and full data storage, it achieves uninterrupted closed-loop detection, fully retaining the core functions of high-accuracy identification of cat-eye targets, low false alarm interference suppression, high-precision positioning and ranging, and continuous and stable detection, ensuring that the detection accuracy and reliability of handheld devices are comparable to those of fixed devices. Unlike fixed devices, this handheld portable system features miniaturized and low-power optimized functional modules: the active illumination module uses a 940nm human eye-safe band miniature semiconductor pulsed laser with a maximum output power optimized to 5W. While effectively exciting cat's eye echo signals and achieving concealed illumination in near and medium-range scenarios, it significantly reduces the size of the laser emission component and the overall power consumption, adapting to the lightweight design requirements of handheld devices. The coaxial imaging acquisition module eliminates the motorized zoom structure, adopting a 100mm fixed-focus telephoto lens with a fixed field of view of 1°. The structure is simpler and more stable, eliminating the need for a zoom drive mechanism and further compressing the optical path volume. At the same time, it can meet the requirements of small field of view and high-resolution target imaging in conventional search scenarios, ensuring clear and complete imaging of cat's eye targets at close range.

[0070] The intelligent vision processing module abandons the large GPU module and adopts the Rockchip RK3588 high-performance, low-power edge computing module, which has lower power consumption, smaller size, and higher integration. It is also equipped with a specially optimized and compressed lightweight cat-eye detection and recognition algorithm model. While maintaining recognition accuracy and anti-interference capability, it significantly reduces the computational load of the algorithm, adapting to the real-time computing needs of embedded devices and ensuring low-latency, high-real-time operation of image analysis, target recognition, and positioning and ranging processes on handheld devices. This embodiment features a highly integrated, one-piece package, reducing the overall size to 120mm × 80mm × 60mm, making it compact and portable. It also has a built-in 10000mAh high-capacity lithium battery, providing ≥4 hours of continuous operation without external power, fully meeting the battery life requirements of long-term field patrols and large-area point-by-point inspections indoors.

[0071] To further simplify the overall structure, reduce the weight and power consumption of the equipment, and adapt to handheld mobile operation scenarios, this portable version removes the servo tracking module. Targeting the manual following and alignment characteristics of handheld detection operations, the automatic servo tracking function has been streamlined. While retaining the core algorithms for target recognition, positioning and ranging, and trajectory output, the detection relies on manual handheld scanning and alignment to complete the investigation. This greatly simplifies the system architecture, reduces equipment cost and overall power consumption, and is more suitable for portable and covert detection applications.

[0072] In summary, this embodiment, while retaining the core innovative principles of active coaxial imaging, intelligent visual recognition, multi-scene adaptive parameter control, and low false alarm rate high-precision detection, achieves the design goals of miniaturization, low power consumption, long battery life, and portability through hardware miniaturization selection, lightweight algorithm deployment, structural simplification and optimization, and power supply system adaptation. In mobile operation scenarios such as individual soldier covert reconnaissance in the field, indoor anti-espionage investigation, and mobile point inspection, it can quickly complete the active detection and intelligent identification of optical targets, accurately identifying optical targets such as hidden cameras, observation mirrors, and photoelectric devices. It possesses advantages such as flexible deployment, no site restrictions, strong concealment, simple operation, and long battery life, effectively compensating for the shortcomings of large fixed detection equipment, such as poor mobility, cumbersome deployment, and inability to conduct mobile investigations. This allows the technical solution of this application to be adapted to both fixed security guard duty and handheld mobile investigation application scenarios, greatly expanding the system's applicability and engineering promotion value.

[0073] In the description of this specification, the references to "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0074] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0075] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0076] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0077] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0078] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. An active imaging system for cat-eye target detection based on intelligent vision, characterized in that, include: The system includes an active illumination module, a coaxial imaging acquisition module, an intelligent vision processing module, an adaptive control module, and a display, control, and storage module; among which, The active illumination module is used to output pulsed laser with adaptively adjustable parameters to actively illuminate the target scene and stimulate the cat's eye effect back reflection echo of the optical device in the target scene. The coaxial imaging acquisition module and the active illumination module adopt a strictly coaxial optical path design for receiving and transmitting back the back reflection echo of the target scene, completing photoelectric conversion and image acquisition, and transmitting the acquired image data to the intelligent vision processing module. The intelligent vision processing module is connected to the coaxial imaging acquisition module, the adaptive control module, and the display, control and storage module respectively. It is used to perform intelligent vision processing on the received image data, complete the identification, positioning and tracking of the cat eye target, and output target information and scene perception data. The adaptive control module is connected to the active lighting module, the coaxial imaging acquisition module, and the intelligent vision processing module respectively. It is used to adaptively adjust the laser output parameters of the active lighting module and the imaging acquisition parameters of the coaxial imaging acquisition module according to the scene perception data and target information, and construct a closed-loop control link. The display, control and storage module is used to realize human-computer interaction, display of real-time imaging images and target information, and storage of raw images, target data and system logs.

2. The active imaging system for cat-eye target detection based on intelligent vision according to claim 1, characterized in that, The active illumination module includes a multi-band laser source, a beam shaping unit, an acousto-optic modulator, and a laser driving circuit; wherein... The multi-band laser source uses a near-infrared pulsed laser, with an output wavelength covering at least one of 808nm, 940nm, and 1064nm, to achieve concealed active illumination. The beam shaping unit includes a collimating lens group and an adjustable beam expander, which are used to collimate and expand the laser beam to achieve adaptive adjustment of the laser divergence angle within the range of 0.1 mrad to 5 mrad. The acousto-optic modulator is used to modulate the laser pulse and output a nanosecond-level narrow pulse laser with an adjustable pulse repetition rate of 10Hz~10kHz. The laser driving circuit is connected to the adaptive control module and is used to receive control commands and adjust the laser's output power, repetition rate, pulse width, and divergence angle.

3. The active imaging system for cat-eye target detection based on intelligent vision according to claim 1, characterized in that, The coaxial imaging acquisition module includes a polarizing beam splitter, a quarter-wave plate, an imaging objective lens group, a narrow-band filter, a planar photodetector, and an image acquisition circuit; wherein... The polarizing beam splitter, in conjunction with the quarter-wave plate, forms a coaxial optical path for transmitting and receiving: the laser output from the active illumination module is reflected by the polarizing beam splitter and then emitted to the target scene through the imaging objective lens group; the back reflection echo from the target scene returns along the original optical path, and after passing through the imaging objective lens group, the polarizing beam splitter, and the quarter-wave plate, it is transmitted and enters the narrowband filter. The center wavelength of the narrowband filter is matched with the laser wavelength of the active illumination module, and the half-width at half-maximum (WHM) is ≤10nm, which is used to filter out stray light from the ambient background. The imaging objective lens group adopts a telephoto zoom objective lens with an adjustable focal length range of 50mm to 500mm and an adjustable field of view range of 0.1° to 5°. The area array photodetector uses a near-infrared enhanced CMOS / CCD / InGaAs detector, supports global exposure, and has an adjustable frame rate range of 1fps to 60fps. The image acquisition circuit is used to perform analog-to-digital conversion and preprocessing of the image, and transmit the acquired image data to the intelligent vision processing module.

4. The active imaging system for cat-eye target detection based on intelligent vision according to claim 1, characterized in that, The intelligent vision processing module adopts an FPGA+ARM heterogeneous computing architecture or an edge GPU computing module, including an image preprocessing unit, a candidate region extraction unit, a deep learning feature extraction and classification unit, a target localization and ranging unit, and a multi-frame correlation tracking unit; wherein... The image preprocessing unit is used to sequentially perform blind pixel correction, non-uniformity correction, adaptive denoising, contrast enhancement, and background difference suppression on the received raw image. The candidate region extraction unit is used to integrate adaptive threshold segmentation connected component analysis with a lightweight target detection network to quickly filter out candidate regions of suspected cat-eye targets and generate candidate target boxes. The deep learning feature extraction and classification unit has a built-in lightweight CNN classification network, which is used to extract features from image slices of candidate regions, distinguish between real cat eye targets and false interference sources, and output the category and corresponding confidence of the cat eye target. The target positioning and ranging unit is used to calculate the pixel coordinates of the target in the image based on the identified cat eye target, obtain the azimuth and pitch angles of the target by combining the imaging intra-participant calibration parameters, calculate the straight-line distance of the target based on the laser time-of-flight (TOF) principle, and output the three-dimensional position information of the target. The multi-frame correlation tracking unit uses an improved DeepSORT multi-target tracking algorithm, combined with Kalman filtering, to perform feature matching and trajectory correlation on cat-eye targets in consecutive frames, and outputs the target's motion trajectory.

5. The active imaging system for cat-eye target detection based on intelligent vision according to claim 4, characterized in that, The training dataset of the lightweight CNN classification and recognition network contains real cat eye target samples and various fake interference samples with different distances, lighting, and backgrounds. The dataset is expanded by data augmentation methods such as rotation, scaling, noise addition, and brightness adjustment. The lightweight CNN classification and recognition network achieves an accuracy rate of ≥99% and a false alarm rate of ≤0.1% for identifying cat-eye targets. The lightweight CNN classification network can identify cat-eye target categories including at least one of camera lenses, telescopes, sniper scopes, drone optoelectronic pods, and vehicle-mounted optical equipment.

6. The active imaging system for cat-eye target detection based on intelligent vision according to claim 1, characterized in that, The adaptive control module includes an environmental perception subunit and a parameter adaptive adjustment subunit; wherein... The environmental perception subunit is used to collect ambient light intensity, image signal-to-noise ratio, and scene background complexity parameters in real time. The parameter adaptive adjustment subunit is used to generate control commands based on environmental perception parameters and target information output by the intelligent vision processing module, and adaptively adjust the laser output power, repetition rate, divergence angle, and pulse width of the active illumination module, as well as the detector exposure time, gain, frame rate, and imaging objective focal length of the coaxial imaging acquisition module.

7. The active imaging system for cat-eye target detection based on intelligent vision according to claim 1, characterized in that, It also includes a servo tracking module; among which, The servo tracking module includes a two-dimensional high-precision servo turntable, a turntable drive circuit, and an encoder; wherein... The servo tracking module is connected to the intelligent vision processing module and is used to receive the azimuth and pitch angle deviation information of the target, drive the two-dimensional servo turntable to rotate in real time, and make the system optical axis continuously aligned with the cat's eye target. The azimuth adjustment range of the two-dimensional servo turntable is ±170°, the pitch adjustment range is -45° to +85°, and the positioning accuracy is ≤0.05°.

8. The active imaging system for cat eye target detection based on intelligent vision according to claim 1, characterized in that, The display, control, and storage module communicates with the intelligent vision processing module and the adaptive control module via gigabit Ethernet; wherein... The touch screen of the display and storage module displays the active imaging image, the bounding box annotation of the cat eye target, the target category, confidence level, azimuth angle, distance, and motion trajectory information in real time; The display, control and storage module provides a human-machine interface that supports switching detection modes, setting parameters, adjusting focus, and inputting target locking commands. The display, control and storage module has a built-in storage unit that supports real-time storage, playback and export of raw image data, target detection data and system operation logs.

9. A method for using an active imaging system for cat-eye target detection based on intelligent vision, as described in any one of claims 1-8, characterized in that, The method includes: S1: The system is powered on and initialized, completing self-tests of each module, parameter calibration, and loading of the deep learning network model. It also receives the user-set detection mode and initial parameters. S2: The adaptive control module sets the initial laser parameters of the active illumination module and the initial imaging parameters of the coaxial imaging acquisition module according to the current ambient light intensity. S3: The active illumination module emits pulsed laser light, which illuminates the target scene through the coaxial optical path of the coaxial imaging acquisition module. The back reflection echo of the target scene returns along the original optical path, is received by the coaxial imaging acquisition module, and the image is acquired and transmitted to the intelligent vision processing module. S4: The intelligent vision processing module performs preprocessing, candidate region extraction, and deep learning classification and recognition on the image data in sequence, distinguishing between real cat eye targets and false interference, and outputting the target's category, confidence level, and coordinate information; it also completes the target's positioning and ranging and multi-frame tracking simultaneously, and outputs the target's three-dimensional position information and motion trajectory. S5: The adaptive control module adjusts the laser parameters of the active lighting module and the imaging parameters of the coaxial imaging acquisition module in real time based on the current scene perception data and target information to form a closed-loop control. S6: The display, control and storage module displays the imaging screen and target information in real time and stores relevant data synchronously; if the tracking mode is enabled, the servo tracking module drives the turntable to continuously lock onto and track the target. S7: Repeat steps S3-S6 until a stop probe command is received.

10. The active imaging method for cat eye target detection based on intelligent vision according to claim 9, characterized in that, In step S5, the adjustment strategy of the adaptive control module includes: When detecting distant targets, increase the laser output power, decrease the laser divergence angle, extend the detector exposure time, and reduce the frame rate; When in a strong light complex background, reduce laser power, increase detector frame rate, and enable multi-frame background difference suppression; When tracking dynamic targets, increase the laser repetition rate and detector frame rate.

Images