Ultraviolet Detection System and Method Based on Spatial Hierarchical Trap-State Modulation Transient Response

Abstract

This invention discloses an ultraviolet detection system and method based on spatially layered trap state modulation transient response. The method includes: designing a detection PIN photodiode with multiple defect states having a spatial gradient along the incident light direction to construct a differential hardware architecture to output a differential ultraviolet photocurrent signal; setting a high-frequency transient sampling window and a conventional timing sampling window; based on the configured time parameters, densely sampling the differential ultraviolet photocurrent signal at a high-frequency sampling rate within the high-frequency transient sampling window to extract ultraviolet band features; within the conventional timing sampling window after the high-frequency transient sampling window ends, continuously sampling the signal at a conventional sampling rate lower than the high-frequency sampling rate to extract time-frequency domain features; and identifying light source events based on the extracted features. This invention solves the problems of low band resolution, susceptibility to background light saturation interference, and excessively high power consumption during high-speed sampling in existing filterless ultraviolet detection technologies.

Description

Technical Field

[0001] This invention belongs to the field of ultraviolet detection technology, and specifically relates to an ultraviolet detection system and method based on spatially layered trap state modulation transient response. Background Technology

[0002] Accurate detection of ultraviolet (UV) radiation is required in modern industrial safety protection (such as welding arc monitoring and high-voltage power grid fault early warning), medical and health care (germicidal lamp dosage monitoring), and daily health management (outdoor sun protection monitoring). Ultraviolet radiation is typically divided into UVA (315-400nm), UVB (280-315nm), and UVC (200-280nm, i.e., the solar blind zone) bands based on wavelength. Among these, UVC and UVB radiation pose a significant threat to the human body, easily causing sunburn.

[0003] Traditional ultraviolet (UV) detection technology faces technological bottlenecks in its evolution towards portable, low-power Internet of Things (IoT) and wearable devices. To achieve detection in specific UV bands, traditional hardware approaches employ expensive UV transmission / visible light cutoff filters (especially UVC filters, which are extremely costly), but these are prone to optical degradation under strong outdoor sunlight. While wide-bandgap semiconductor (such as SiC and AlGaN) UV detectors can only respond to UVC while blocking UVA and UVB, the heterogeneous integration of wide-bandgap semiconductors with mature Si CMOS chip circuits presents challenges, resulting in high detector costs and hindering integration into the wearable device ecosystem. For example, Chinese patent CN109585592B discloses a p-BN / i-AlGaN / n-AlGaN ultraviolet detector and its fabrication method. From bottom to top, it includes: a substrate, an AlN nucleation layer, an AlN intrinsic layer, an AlGaN intrinsic layer, an n-type AlGaN layer, an n-electrode, an i-type AlGaN layer, a p-type BN layer, and a p-electrode. The p-type BN layer uses Mg-doped wurtzite boron nitride material with a thickness of 60-100 nm and a doping concentration of 5 × 10⁻⁶. 17 -1×10 19 cm -3 The above-mentioned Mg-doped wurtzite boron nitride material. This invention can only detect UVC and part of UVB, and cannot cover and identify signals in the entire ultraviolet band.

[0004] To achieve filterless ultraviolet detection, a novel technical approach based on the transient impulse response of photodetectors for filterless band classification has emerged. The basic principle is to utilize the differences in penetration depth of different wavelengths of light in silicon (UVC is extremely shallow, visible light is deeper), and to infer the band by analyzing the time constant of the transient rise or fall curve of the photocurrent. However, existing band classification based on transient impulse response has a fatal flaw: extremely low band identification accuracy. The resolution threshold between adjacent bands is typically around 100 nm, making it difficult to stably distinguish between UVA, UVB, and UVC signals with closely adjacent spectra and narrow bandwidths in complex environments. The fundamental physical reason for this flaw lies in the fact that existing transient response band classification techniques rely on the inherent defects and surface states of the detector, without designing for defect distribution. These inherent defects and surface states can only distinguish between ultraviolet light (surface recombination) and visible light (bulk recombination) with drastically different transient responses. In existing technologies, the surface recombination time constant corresponding to UVC absorption and the shallow bulk recombination time constant corresponding to UVB absorption severely overlap, failing to provide a sufficiently sharp physical boundary. When UVC and UVB irradiate simultaneously, the transient current waveforms become severely aliased, making it impossible for backend algorithms to extract high-confidence band classification results from the blurred signals at the physical source. Furthermore, current technology does not consider that ambient visible light background current can overwhelm weak ultraviolet current signals, resulting in a loss of ultraviolet spectrum classification capability at the source.

[0005] On the other hand, existing ultraviolet detection technologies have not undergone in-depth time-frequency characteristic optimization for complex "ultraviolet events." For example, ultraviolet arc discharge is accompanied by the nonlinear collapse and reignition process of plasma, and its transient ultraviolet radiation intensity exhibits extremely irregular jumps and spikes on the microsecond to millisecond scale, with its ultraviolet signal possessing unique high-frequency chaotic white noise; while artificial ultraviolet lamps exhibit periodic light intensity flicker of 100Hz or 120Hz, unique to being driven by AC power grids. Under the current technological framework, in order to capture the high-frequency characteristics of arcs and other ultraviolet events, the detector's driving circuit often needs to maintain an extremely high ADC sampling rate globally (such as at the megahertz level). This not only generates massive data throughput but also causes the power consumption of the microcontroller (MCU) to soar, making it unsuitable for wearable devices with limited battery capacity, such as smart glasses and smartwatches. Summary of the Invention

[0006] The purpose of this invention is to provide an ultraviolet detection system and method based on spatially layered trap state modulation transient response, in order to solve the application pain points of existing filterless ultraviolet detection technology, such as low band resolution, susceptibility to background light saturation interference, and excessive power consumption during high-speed sampling. This invention provides a low-cost, low-power, and high-precision solution for industrial safety and consumer-grade health monitoring.

[0007] The technical solution of the present invention is as follows: On one hand, the present invention provides an ultraviolet detection system based on spatially hierarchical trap state modulation transient response, comprising: A differential hardware architecture is used to output differential ultraviolet photocurrent signals. The differential hardware architecture is based on a detection PIN photodiode, which has a multi-layer defect state design with a spatial gradient along the incident light direction. This design is used to modulate the transient response time constants of photogenerated carriers in the UVC, UVB, and UVA bands at the physical layer, so that the dynamic characteristics of photogenerated carriers in different bands can be separated in a stepwise manner on the time axis. The feature extraction module performs intensive sampling of the differential ultraviolet photocurrent signal at a high-frequency sampling rate within the high-frequency transient sampling window that enters the transient feature extraction cycle, and extracts ultraviolet band features including UVC, UVB, and UVA band features; within the regular time-series sampling window after the high-frequency transient sampling window ends, the signal is continuously sampled at a regular sampling rate lower than the high-frequency sampling rate to extract time-frequency domain features. A multi-dimensional decision engine identifies light source events based on the ultraviolet band features and time-frequency domain features.

[0008] Preferably, the differential hardware architecture is as follows: The differential hardware architecture also includes a reference PIN photodiode, which is manufactured on the same substrate as the probe PIN photodiode and has a common spatial layered trap state design. The surface of the reference PIN photodiode is covered with an ultraviolet cutoff filter layer. The probe PIN photodiode is in an exposed state. The output terminals of the detection PIN photodiode and the reference PIN photodiode are connected to the differential front-end circuit. The differential operation cancels the visible light and near-infrared background photocurrent and outputs a differential ultraviolet photocurrent signal to the feature extraction module.

[0009] Preferably, the detection PIN photodiode adopts a P++ / P+ / N- / N+ doping profile design with spatial gradient, and the specific structure includes: A shallow defect structure layer, corresponding to the physical absorption region of the UVC band, introduces a high concentration of surface defect states on the surface of the P++ doped material layer through a plasma treatment process to capture UVC photogenerated carriers with extremely shallow penetration depth. The intermediate defect structure layer corresponds to the physical absorption region of the UVB band. Medium-deep energy level traps are controllably introduced into the P+ doped material layer through a light mass ion implantation process to modulate the SRH recombination and release time of photogenerated carriers in the UVB band. The deep bulk region structure corresponds to the physical absorption region of the UVA band. The N-lightly doped layer and the N+ doped layer are intrinsic epitaxial defects. During detection, the N-lightly doped layer is not completely connected. UVA photogenerated carriers are absorbed in the deep undepleted bulk region of the N-lightly doped layer and the N+ doped layer and diffuse to the depletion region.

[0010] Preferably, the time parameter for the feature extraction module to extract ultraviolet band features within the high-frequency transient sampling window is configured to map to the time constant of the detection PIN photodiode physical structure: The highest peak value of current overshoot in the 0 to 20 microsecond range is extracted and mapped to the UVC band characteristics dominated by the shallow defect structure layer. Extract the integral area of ​​the current curve bulge in the 50-100 microsecond interval to map the UVB band characteristics dominated by the middle defect structure layer. The average slope of the monotonically rising current in the range of 200 microseconds to 1000 microseconds is extracted to map the UVA band characteristics dominated by the deep body structure layer.

[0011] Preferably, the extraction of time-frequency spectral features in the feature extraction module specifically involves: calculating the peak factor of the signal within the conventional time-series sampling window, and performing a fast Fourier transform to extract the periodic flicker features in the frequency domain to obtain a spectrum.

[0012] Preferably, the multi-dimensional decision engine includes a logic for determining dangerous arc events: when the UVC band feature in the ultraviolet band is extracted in the high-frequency transient sampling window, and the peak factor calculated in the conventional time-series sampling window is greater than or equal to a preset chaos threshold, it is determined that an arc discharge event exists in the environment.

[0013] Preferably, the multi-dimensional decision engine further includes the judgment logic for AC-driven ultraviolet lamp events: when any one or more band features in the ultraviolet band features are extracted in the high-frequency transient sampling window, and the peak factor calculated in the conventional time-series sampling window is less than a preset chaos threshold, and the spectrum map has a main energy column and a decreasing harmonic column at the AC power grid fundamental frequency and its integer multiples, it is determined that there is an AC-driven ultraviolet light source event in the environment.

[0014] Preferably, the multidimensional decision engine also includes logic for calculating ultraviolet radiation power: when the multidimensional decision engine does not trigger the arc discharge event and the AC-driven ultraviolet lamp event determination, the system performs statistical averaging on the ultraviolet band features extracted by the feature extraction module, and inversely calculates the absolute radiation power values ​​of UVC, UVB, and UVA under the current environment based on a preset discrete mapping table or fitting curve.

[0015] Preferably, the system further includes a driving and timing control module, which is configured to operate in a periodic pulse trigger mode, controlling the detection PIN photodiode to alternately be in a sleep release cycle and a transient feature extraction cycle on the time axis; in the sleep release cycle, the system is powered off to clear the accumulated charge of the multilayer defect states inside the detection PIN photodiode, and the trap energy level is forced to recover to the zero potential energy baseline.

[0016] On the other hand, the present invention provides an ultraviolet detection method based on spatially hierarchical trap state modulation transient response, comprising the following steps: Design a detection PIN photodiode with a multi-layer defect state having a spatial gradient along the incident light direction, and construct a differential hardware architecture based on the detection PIN photodiode to output a differential ultraviolet photocurrent signal; A high-frequency transient sampling window and a regular timing sampling window are set, and the time parameter for extracting features within the high-frequency transient sampling window is configured to map to the time constant of the physical structure of the PIN photodiode being detected; The acquisition of ultraviolet band features includes: extracting UVC, UVB, and UVA band features respectively in a high-frequency transient sampling window based on configured time parameters; and continuously sampling the signal at a conventional sampling rate lower than the high-frequency sampling rate within a conventional time-series sampling window to extract time-frequency domain features. Light source event identification is performed based on the ultraviolet band features and time-frequency domain features.

[0017] In another aspect, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any embodiment of the present invention.

[0018] Compared with the prior art, the present invention has the following technical effects: 1. Significantly Improved Band Recognition Accuracy: Through the trap-layer design of the detector, this invention transforms the transient response times of UVC, UVB, and UVA from the natural "fuzzy overlap" to a "stepped layering." The combination of plasma-damaged surface defects and lightweight ion implantation of mid-layer defects makes the transient pulse characteristics of photogenerated carriers in each band clearly defined, effectively improving the accuracy of ultraviolet band classification. Furthermore, this invention introduces light source characteristics (arc plasma chaos / power grid flicker) to correspond to UVC / UVB / UVA events in real-life or industrial scenarios. On the one hand, it enhances the accuracy of ultraviolet band classification under complex lighting backgrounds by leveraging the identified ultraviolet light source characteristics; on the other hand, the combination of ultraviolet band classification and ultraviolet light source identification provides event alerts to users, reducing the probability of false alarms.

[0019] 2. Efficient Inversion and Quantitative Monitoring of Multi-Band Ultraviolet Power: Building upon qualitative band classification, this invention further establishes a quantitative mapping relationship between transient physical characteristics and absolute radiative power. Through a pre-calibrated fitting mapping, the system accurately inverts the UVC overshoot peak value, UVB curve bulge area, and UVA rise slope characteristics extracted from the high-frequency transient window into the corresponding absolute radiative power values ​​for the respective bands. This feature enables the system not only to "identify" ultraviolet radiation but also to perform long-term radiation dose integration calculations, showing broad application prospects in quantitative detection scenarios such as medical sterilization dose monitoring and human sun protection safety early warning.

[0020] 3. Extremely low power consumption and computational efficiency: The system described in this invention operates at a high frequency of 500kHz for only 1 millisecond within 1 minute, and spends the rest of the time in low-frequency sampling at 10kHz or in deep sleep. This asymmetric timing, combined with the underlying multi-layer defect time constant, eliminates the need for complex algorithms. It can identify UVA / UVB / UVC events using only basic peak extraction, integration (area), differentiation (slope), and one-dimensional FFT operations. The entire logic can be implemented using a very low-cost ordinary microcontroller (MCU), keeping the overall system power consumption at the microwatt level.

[0021] 4. Hardware BOM costs decrease exponentially: Existing silicon detectors require expensive UVC filters to identify UVC, and for full-band UV identification and classification, three filters (UVC, UVB, and UVA) are needed. This invention requires only one detection silicon PIN photodiode, one reference silicon PIN photodiode (with inexpensive UV cutoff resin), and a low-cost ordinary MCU controller, effectively reducing the application cost of full-band UV detectors. Attached Figure Description

[0022] Figure 1 This is an overall structural diagram of the ultraviolet detection system based on spatially hierarchical trap state modulation transient response as described in this invention; Figure 2 This is a schematic diagram of the transient response of UVC, UVB, and UVA incident individually on the detection PIN photodiode. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments of the present application and with reference to the accompanying drawings.

[0024] Example 1 This embodiment provides an ultraviolet detection system based on spatially hierarchical trap state modulation transient response, such as... Figure 1 As shown, it includes: A differential hardware architecture is used to output differential ultraviolet photocurrent signals. The differential hardware architecture is based on a probe PIN photodiode, which has a multi-layer defect state design with a spatial gradient along the incident light direction. This design is used to modulate the transient response time constants of photogenerated carriers in the UVC, UVB, and UVA bands at the physical layer, so that the dynamic characteristics of photogenerated carriers in different bands can be separated in a stepwise manner on the time axis.

[0025] The feature extraction module performs intensive sampling of the differential ultraviolet photocurrent signal at a high-frequency sampling rate within the high-frequency transient sampling window that enters the transient feature extraction cycle, extracting ultraviolet band features including UVC, UVB, and UVA band features; within the regular time-series sampling window after the high-frequency transient sampling window ends, the signal is continuously sampled at a regular sampling rate lower than the high-frequency sampling rate to extract time-frequency domain features.

[0026] A multi-dimensional decision engine identifies light source events based on the ultraviolet band features and time-frequency domain features.

[0027] As a preferred embodiment of this invention, the differential hardware architecture is used to address the issues of filter cost and visible light saturation. The specific structure is as follows: The differential hardware architecture also includes a reference PIN photodiode, which is manufactured on the same substrate as the probe PIN photodiode and has a common spatial layered trap state design. The surface of the reference PIN photodiode is covered with an ultraviolet cutoff filter layer, while the probe PIN photodiode is in an exposed state.

[0028] The output terminals of the detection PIN photodiode and the reference PIN photodiode are connected to the differential front-end circuit. The differential operation cancels the visible light and near-infrared background photocurrent and outputs a differential ultraviolet photocurrent signal to the feature extraction module.

[0029] The probe PIN photodiode contains no filter and receives the full spectrum; the reference PIN photodiode has an ultraviolet cutoff filter of "ultraviolet absorbing resin + blue glass" covering its surface, allowing it to receive only visible and near-infrared light. The probe and reference PIN photodiodes, through a subtraction circuit in an analog front end, remove background photocurrent from ambient visible and near-infrared light, directly outputting a differential ultraviolet AC signal.

[0030] As a preferred embodiment of this invention, this invention addresses the problems of low band classification accuracy and band aliasing by designing defects and doping at the device level for full-band ultraviolet detection (UVA, UVB, UVC). The detection unit employs a gradient-varying P++ / P+ / N- / N+ doping design. To achieve precise physical partitioning of space traps, the device undergoes the following trap engineering treatment, specifically including: A shallow defect layer, corresponding to the physical absorption region of the UVC band, is constructed by introducing a high concentration of surface defect states on the surface of the P++ doped material layer through a plasma treatment process to capture UVC photogenerated carriers with extremely shallow penetration depth. In this P++ region, the hole concentration is controlled to be 10-1. 18 cm -3 The magnitude of the signal anchors the physical boundary of the PN junction depletion region and generates an extremely strong built-in electric field. Photogenerated carriers, generated by ultra-shallow (<20nm) UVC photons, undergo intense capture and escape interactions with high-concentration surface defects under a strong electric field, producing an extremely rapid and sharp displacement current overshoot. By controlling the surface defect energy levels and density of states through plasma damage processes, the rise time of the transient current overshoot can be controlled within 10-20μs to facilitate sampling by low-power circuits.

[0031] A mid-defect structure layer, corresponding to the physical absorption region of the UVB band, is constructed by controllably introducing mid-deep level traps into the P+ doped material layer using a light-mass ion implantation process. These traps are used to modulate the SRH recombination and release times of photogenerated carriers in the UVB band. This depth corresponds to the main penetration absorption region of UVB photons (50-100 nm). By forming specific deep-level bulk defects through a suitable hydrogen implantation process, the SRH recombination time of minority carriers in the region is controlled. The carrier response time excited by UVB is controllably designed within the 50-100 μs range, thus forming a significant physical distinction from the extremely fast response of UVC.

[0032] The deep bulk region structure corresponds to the physical absorption region of the UVA band. The N-lightly doped and N+ doped layers are intrinsic epitaxial defects. During detection, the N-lightly doped layer is not completely penetrated. UVA photogenerated carriers are absorbed in the deep undepleted bulk region of the N-lightly doped and N+ doped layers and diffuse to the depletion region. The deep N-lightly doped and N+ doped layers do not intentionally introduce doping, maintaining an extremely low background defect density. UVA photons penetrating to this depth (>100nm or even micrometers) achieve extremely long recombination lifetimes for photogenerated minority carriers due to the extremely low defect trapping rate in this region. Based on the common background defect density of single-crystal silicon N-lightly doped layers, this long-distance diffusion process generally has a slow current rise slope of 200-1000μs, thus forming a clear physical distinction from the transient currents of UVB and UVC.

[0033] The aforementioned trap engineering improves the transient current difference of UVB / UVC / UVA incident light pulses, significantly enhancing the spectral resolution of filterless band classification technology.

[0034] In a preferred embodiment of this invention, the time parameter for the feature extraction module to extract ultraviolet band features within the high-frequency transient sampling window is configured to map to the time constant of the detection PIN photodiode physical structure: The highest peak value of current overshoot in the 0 to 20 microsecond range is extracted and mapped to the UVC band characteristics dominated by the shallow defect structure layer.

[0035] The area of ​​the current curve bulge within the 50-100 microsecond interval is extracted to map the UVB band characteristics dominated by the middle defect structure layer.

[0036] The average slope of the monotonically rising current in the range of 200 microseconds to 1000 microseconds is extracted to map the UVA band characteristics dominated by the deep body structure layer.

[0037] It is worth noting that the 10-20μs, 50-100μs, and 200-1000μs time features extracted within the high-frequency transient sampling window in this invention are not arbitrarily set algorithm parameters, but are calculated based on the penetration depth of photons in silicon-based materials at different wavelengths and the corresponding spatial positions, representing a physically feasible defect control engineering approach. The specific calculation process is as follows: The transient overshoot time in the UVC band is set to the 10-20 μs range: The UVC band (e.g., 254 nm) has extremely shallow penetration depth, and almost all of it is absorbed in the P++ surface layer. Taking the surface defect state with an activation energy of 0.41 eV introduced by plasma damage as an example, according to the semiconductor Poisson equation and the formula for the trap emission time constant... Substitute the effective density of states of the conduction band of room temperature silicon. Take as Electron thermal motion velocity Take as Typical electron trapping cross section of surface defect state Take as ; Activate the trap; Boltzmann's constant; Thermodynamic temperature is taken as room temperature (300K); the carrier emission time constant can be obtained. ;when hour, Calculations show that the physical duration of the UVC carrier release characteristic dominated by shallow traps can fall between 10 and 20 microseconds. This forms the underlying hardware basis for the 500kHz (2μs interval) high-frequency sampling of this invention, ensuring that high-fidelity data points can be stably acquired for each UVC current overshoot.

[0038] The transient response time for the UVB band is set to the 50-100μs range: The penetration depth of the UVB band is approximately 50-100 nm, and photogenerated carriers are mainly generated in the P+ layer. Here, hydrogen ion injection forms a mid-to-deep level trap, altering the SRH trapping time of minority carriers, as shown in the formula. Based on the hydrogen ion implantation process configuration, the volume defect density N introduced in this region is... t Regulated The magnitude, its typical hole capture cross section The physical value range is usually in to Between these, the thermal motion rate of holes in silicon at room temperature Take as Corresponding It falls within the range of 50μs to 100μs.

[0039] Theoretical calculations show that deeper trap energy levels require a longer thermal excitation time for carriers to be released after being trapped, thus significantly broadening the rising edge of the photocurrent and forming a characteristic raised integral area. The time span of this area can be tuned to the range of 50 μs to 100 μs through trap engineering.

[0040] The pulse rise time for the UVA band is set to the 200-1000μs range: The UVA band (e.g., 365 nm) has the deepest photon penetration depth (greater than 100 nm, even reaching the micrometer level), where most photogenerated carriers are excited in the deep, unconnected N- and N+ doped regions. These unconnected regions lack a strong built-in electric field, and photogenerated minority carriers diffuse to the depletion region via their concentration gradient. During this diffusion process, the time for the photocurrent to rise to steady state is limited by the minority carrier recombination lifetime in this region. Since the lightly doped N- and N+ doped layers do not introduce artificial defects, the residual intrinsic deep-level defect density... Controlled at an extremely low level (taken as) to The range corresponds to ordinary commercial silicon wafers. According to the SRH recombination theory, the intrinsic lifetime of minority carriers in silicon is given by the formula... Decision. Typical hole trapping cross section of residual deep-level recombination centers in silicon at room temperature. Approximately Hole thermal motion speed Take as From this, the carrier lifetime can be calculated. In the range of 200μs to 1000μs.

[0041] The extremely low background defect density of Si materials endows photogenerated minority carriers with an extremely long diffusion time. Theoretical calculations show that, limited by the photocurrent diffusion hysteresis characteristic of minority carrier lifetime (corresponding to a slow, monotonically increasing ramp), its time span can be designed to be in the range of 200 μs to 1000 μs.

[0042] As a preferred embodiment of this example, the extraction of time-frequency spectral features in the feature extraction module specifically involves: calculating the peak factor of the signal within the conventional time-series sampling window, and performing a fast Fourier transform to extract the periodic flicker features in the frequency domain to obtain a spectrum.

[0043] The purpose of calculating the crest factor is to determine the characteristics of the light source. Considering the temporal characteristics of ultraviolet light sources, such as UVC from arc discharge, the transient ultraviolet radiation intensity exhibits extremely irregular jumps and spikes on the microsecond to millisecond scale, which in signal science manifests as high dynamic range random noise characteristics, i.e., extremely high signal chaos. This signal characteristic can be comprehensively extracted using the asymmetric dual-speed sampling strategy of this invention and quantified by the crest factor, which characterizes the signal chaos. The crest factor conforms to the engineering definition and is configured as: the ratio of the global maximum absolute peak value of the discrete ultraviolet photocurrent signal within a single wake-up cycle to the effective value (RMS) of the signal within that complete cycle, used to quantitatively characterize the degree of peak distortion of the ultraviolet light intensity time-domain waveform. Because the high-frequency transient sampling window can accurately capture the microsecond-level extremely high energy spikes generated by arc plasma instability, and the subsequent extended conventional time-series sampling window can provide a sufficiently long time baseline to calculate a stable effective value, the peak factor calculated by combining the two will be significantly greater than that of conventional steady-current ultraviolet lamps or natural light with smooth envelopes, thereby identifying ultraviolet arc events.

[0044] Within a single wake-up cycle (e.g., 200 milliseconds), the system employs an asymmetric sampling strategy of "tight at the beginning and loose at the end," which self-consistently divides the time axis into two sampling windows with independent functions: High-frequency transient sampling window (band classification region, e.g., 0-1ms): Initiates dense high-frequency sampling at 500kHz to capture microsecond-level physical layer trap dynamics. 500kHz corresponds to a 2μs sampling interval, which can effectively resolve UVC transient photocurrent overshoot caused by surface defects (10-20μs, yielding 5-10 high-fidelity data points), UVB photocurrent rise caused by mid-layer defects (50-100μs), and the slow increase in photocurrent caused by deep UVA absorption and minority carrier diffusion (200μs-1ms).

[0045] The standard timing sampling window (light source characterization region, e.g., 1-200ms): Frequency is downgraded to 10kHz for continuous sampling, extracting the effective value of the signal within this period and performing FFT to extract periodic flicker features. This 10kHz sampling rate is based on the fact that for common arc discharges, the upper limit of the chaotic white noise bandwidth of macroscopic plasma fluctuations is typically around 5kHz. According to the Nyquist sampling theorem, a 10kHz sampling rate can completely preserve the chaotic fluctuation energy within this frequency band, thus accurately calculating the true effective value baseline of the arc signal. Combined with the maximum absolute peak value captured by the high-frequency transient sampling window, both ensure the rigor and accuracy of the crest factor calculation. Simultaneously, this sampling rate is backward compatible with the extraction of flicker features from 100Hz / 120Hz AC UV lamps.

[0046] As a preferred embodiment of this invention, the multi-dimensional decision engine introduces a light source feature dimension to enhance the confidence level of ultraviolet (UV) identification. Common UVC events in industrial and residential environments include electric arc discharge and deep UV lamps, while common UVB events include various UV lamp sources and solar irradiation.

[0047] In the calculation of the number of peaks in the feature extraction module, this invention defines a peak factor of a specific value (e.g., preferably ≥5) as the chaos threshold for determining the nonlinear jump characteristics of the arc. When the peak factor calculated by the system reaches or exceeds this threshold, it is determined that a high-chaos arc discharge event has been captured, thereby achieving accurate differentiation between arc characteristics and conventional ambient light sources and solving the false alarm problem in complex environments.

[0048] The multi-dimensional decision engine includes logic for determining dangerous arc events: when UVC band features are extracted from the ultraviolet band features in the high-frequency transient sampling window, and the peak factor calculated in the conventional time-series sampling window is greater than or equal to a preset chaos threshold, an arc discharge event is determined to exist in the environment. Specifically, if there is a transient photocurrent overshoot (preferably 10-20 μs, identified as UVC) in the high-frequency transient sampling window, and the peak factor calculated in the conventional time-series sampling window is ≥ n (which can be set based on actual application scenario requirements, usually set to 5), it indicates the presence of a severe spike in plasma chaotic discharge, and an arc event is determined to exist in the environment. The system issues a UVC alarm and indicates a potential arc discharge event.

[0049] As a preferred embodiment of this example, the multi-dimensional decision engine further includes the judgment logic for AC-driven ultraviolet lamp events: when any one or more band features in the ultraviolet band features are extracted in the high-frequency transient sampling window, and the peak factor calculated in the conventional time-series sampling window is less than a preset chaos threshold, and the spectrum map has a main energy column and a decreasing harmonic column at the AC power grid fundamental frequency and its integer multiples, it is determined that there is an AC-driven ultraviolet light source event in the environment. Specifically, if the high-frequency transient sampling window exhibits transient photocurrent overshoot (10-20µs, identified as UVC), photocurrent rise (50-100µs, identified as UVB), or slow photocurrent increase (200µs-1ms, identified as UVA), but the peak factor calculated by the conventional time-series sampling window is <5, and the FFT extracts periodic flicker features showing a huge energy column at 100Hz / 120Hz (the frequency can be set based on the actual application scenario requirements; AC power grid frequency harmonics are generally 100Hz / 120Hz) and its harmonics, it indicates the presence of an AC power grid-driven ultraviolet light source in the environment. The system then indicates potential UVC, UVB, and UVA ultraviolet lamp irradiation. Considering the UVC irradiation wavelength of excimer lamps (222nm), low-pressure mercury lamps (254nm), medical UVB lamps (311nm), and UVA curing lamps (365nm), in real-world scenarios, there is generally only a single ultraviolet light source. Therefore, the AC ultraviolet lamp identification method of this invention has good confidence.

[0050] As a preferred embodiment of this example, the multidimensional decision engine also includes the logic for calculating ultraviolet radiation power: when the multidimensional decision engine does not trigger the determination of arc discharge event and AC-driven ultraviolet lamp event, the system performs statistical averaging on the ultraviolet band features extracted by the feature extraction module, and inversely calculates the absolute radiation power values ​​of UVC, UVB, and UVA under the current environment based on the preset discrete mapping table or fitting curve.

[0051] Based on the system's periodic sleep reset and wake-up mechanism, the detector undergoes a transient process from zero to steady state each time it is powered on. It is worth noting that the absolute radiative power values ​​of UVC, UVB, and UVA in the current environment are... Figure 2 The UVC transient photocurrent overshoot peak value, the UVB photocurrent curve bulge area, and the UVA photocurrent rise slope shown exhibit a linear or quasi-linear correspondence: UVC band and peak overshoot: Almost all UVC photons are absorbed on the surface. The greater the UVC radiation power, the higher the concentration of instantaneously excited surface photogenerated carriers. These carriers are instantaneously separated under a strong electric field and captured / released by shallow traps, resulting in a higher peak overshoot of the displacement current.

[0052] UVB band and curve bulge area: UVB photons penetrate to the middle layer and interact with mid-to-deep energy level traps. The greater the UVB power, the greater the total number of charge carriers participating in SRH recombination and release. On the current-time curve, the current integral of this delayed release (i.e., the curve bulge area) must be larger.

[0053] UVA band and rising slope: UVA photons are absorbed in the deep bulk region, and photogenerated carriers diffuse to the depletion region. The greater the UVA power, the higher the minority carrier generation rate in the deep bulk region, and the faster and larger the absolute gradient of the spatial concentration gradient between the bulk region and the depletion region boundary is established. According to Fick's diffusion law, the larger the concentration gradient, the greater the instantaneous diffusion current. Therefore, the average slope of the monotonically rising photocurrent with time increases accordingly.

[0054] In a preferred embodiment, the system further includes a driving and timing control module configured to operate in a periodic pulse trigger mode, controlling the detection PIN photodiode to alternately operate in a sleep release cycle and a transient feature extraction cycle on the time axis. During the sleep release cycle, the system is powered off to clear the accumulated charge of the multilayer defect states inside the detection PIN photodiode, forcing the trap energy level to return to the zero potential energy baseline, thereby solving the background light saturation and power consumption problems. The system operates with an extremely low duty cycle, for example, waking up for only 200 milliseconds every minute before entering a power-off sleep state (T). off Actively clearing the trap charge accumulation caused by ambient background light, forcing the detector's trap energy level state to return to the baseline level, ensuring that the multilayer defects have a repeatable transient response to newly incident ultraviolet photons each time it is woken up. Power-off sleep time T off The carrier thermal release time constant is set to be strictly greater than that of the designed mid-level energy trap.

[0055] The following are some specific examples: Example 1: Industrial arc detection and alarm system integrated into smart glasses Application Scenario Background: Industrial production workshops are characterized by dense wiring, intense AC lighting (such as high-pressure mercury lamps and LED lights), and frequent direct sunlight streaming through skylights. When an electrical short circuit occurs in the workshop, it can generate a dangerous electric arc rich in the UVC band. The system described in this invention can be embedded in a smart eyeglass frame to help inspection engineers monitor for electric arc events.

[0056] System workflow and implementation details: PIN photodiode design for detection: The PIN photodiode for detection is designed with P++ / P+ / N- / N+ doping, and the doping concentration and trap design are shown in Table 1. The module uses the same process to fabricate two PIN devices on a silicon wafer: one device uses the standard PIN photodiode process, and the other, a detector package, has a corresponding lens containing a UV-absorbing resin and a blue glass composite filter.

[0057] Table 1 Design Parameters of Detection PIN Photodiode

[0058] Hardware Integration and Trap Reset: A tiny differential PIN detector module is integrated into the frame of the smart glasses. The smart glasses' MCU controls the detector to wake up every minute. During a 60-second power outage, the deep / shallow energy level traps at the P++ layer and P+ / N- interface of the silicon-based material release all electrons excited by the intense workshop light, and the detector returns to its baseline detection state.

[0059] 1ms Burst Sampling (Precise Initial Screening): At the moment of wake-up, weak arc photons from a distance reach the exposed probe PIN photodiode. Simultaneously, background visible light reaches both the probe PIN photodiode and the reference PIN photodiode. The two diodes are differentially analyzed at the analog front end to extract the ultraviolet signal from the probe PIN photodiode. The high-speed ADC inside the MCU continuously captures 500 points at a rate of 500kHz (taking 1ms). Because the arc is rich in high-energy UVC photons, its penetration depth in silicon is extremely shallow (<10nm).

[0060] Underlying physical process: UVC photons are absorbed in the P++ layer. This layer, having undergone plasma damage treatment, has a high concentration of surface defects, and is 1×10-1 18 cm -3 The high hole concentration anchors an extremely strong built-in electric field. Photogenerated carriers are instantly and forcefully separated and rapidly captured / released by defects, generating a huge displacement current.

[0061] Algorithm action: The MCU algorithm detects a sharp transient current overshoot in the range of 10μs to 20μs, confirms the presence of UVC components, and extracts the maximum peak current.

[0062] 199ms Standard Sampling (Arc Detection): After 1ms, the ADC frequency is reduced to 10kHz, and 1990 points are collected again. The MCU calculates the effective value (RMS) of this 199ms data sequence. Since an electric arc is essentially a high-frequency chaotic discharge of plasma, its intensity sequence is full of nonlinear spikes and glitches. The MCU calculates that the crest factor is as high as 8.5, which is much greater than the arc event detection threshold of 5.

[0063] Decision and Execution: Decision tree logic gate trigger: [UVC overshoot == True] && [crest factor >= 5]. The system confirms it as a dangerous arc, not a workshop lighting lamp (whose crest factor is usually less than 2), and triggers an arc alarm.

[0064] Example 2: UVB / UVA component monitoring and sun protection early warning system integrated into outdoor smart portable devices Application Scenario Background: When users wear smart portable devices (such as smartwatches) for outdoor activities, they are exposed to strong sunlight. After sunlight passes through the ozone layer, the UVC band is completely absorbed, and the main UVA rays reaching the ground (causing skin aging and darkening) and a small amount of UVB rays (causing skin redness and sunburn) are UVA rays. Outdoor protection is necessary by obtaining the current UVB and UVA doses of the environment.

[0065] System workflow and implementation details: PIN detection module design: The PIN detection module design in Example 1 is adopted.

[0066] Asymmetric sampling and physical feature mapping: The system wakes up to sample for 200 milliseconds every minute.

[0067] UVC: During a 500kHz burst sampling of 0-1ms, the MCU did not detect an extreme overshoot of 10-20μs (UVC does not exist in nature, and the P++ layer was not excited).

[0068] UVB: The underlying physical processes include: UVB photons penetrate to a depth of approximately 10-100 nm, reaching the P+ / N- interface. Mid-level traps are distributed throughout the P+ layer. The SRH recombination rate of carriers at this location is artificially modulated, resulting in a broadened UVB response time. The algorithmic processes include: The MCU detects a moderately wide waveform bulge within the 50μs to 100μs range; the MCU integrates this bulge region to extract the area feature value S. UVB .

[0069] UVA: The underlying physical processes include: UVA photons penetrate to a depth >100nm, entering the high-purity N- / N+ deep bulk region. There are no artificial defects here; charge carriers move slowly via pure diffusion and drift. The algorithmic processes include: the MCU detects a gently rising slope within the 200μs to 1000μs range and calculates the average slope K of this slope. UVA .

[0070] Interference elimination: In the subsequent 10kHz regular sampling, the MCU calculated the crest factor to be approximately 1.4 (extremely low), and after performing the FFT transformation, the spectrum showed no energy bars at 100Hz / 120Hz (natural light has no flicker).

[0071] Decision-making and power inversion: Decision tree logic gate trigger: [No UVC]&&[Crest factor < 5]&&[FFT has no 100Hz / 120Hz features]. The system determines that the current environment is either pure natural sunlight or steady-current ultraviolet radiation. The MCU extracts the S within the entire 1-millisecond high-frequency burst sampling window. UVB and K UVA The extracted area features (UVB) and slope features (UVA) are substituted into a multivariate linear regression equation pre-calibrated based on a discrete mapping table. The smart portable device displays the currently calculated UVA and UVB power in real time and triggers a sun protection warning system to remind the user to take ultraviolet protection.

[0072] Example 3: Portable medical UVC germicidal lamp dosage monitor Application Scenario Background: In the ultraviolet disinfection process in hospital operating rooms or laboratories, low-pressure mercury lamps (pure UVC) at 254nm are typically used. However, as the usage time of mercury lamps increases, the intensity of their UVC radiation decreases, while the visible light (blue light) portion may still be bright, making it impossible for the human eye to judge their bactericidal efficacy.

[0073] System workflow and implementation details: Detection of PIN photodiode: The differential design of the dual-PIN photodiode in Example 1 is adopted.

[0074] Dual verification of band and light source: The monitoring instrument was placed in a sterilization chamber. After wake-up, a 500kHz burst sampling captured an extremely sharp overshoot peak within 10-20μs, confirming the presence of real high-energy UVC photons (triggering plasma damage defects in the P++ layer). Confirmation of light source properties: After collecting data using 10kHz routine sampling, the peak factor was calculated to be less than 5 (excluding arcing). Subsequent FFT analysis clearly showed an extremely large and sharp energy column at the 100Hz harmonic harmonic of the AC power grid on the spectrum, along with decreasing harmonic columns at 200Hz and 300Hz.

[0075] Decision-making and dosage accumulation: Decision tree logic gate trigger: [UVC overshoot == True] && [Crest factor < 5] && [FFT shows 100Hz main energy]. The system confirms that the AC UVC germicidal lamp is working normally. Further actions include: The MCU extracts the highest peak value of the 10-20μs overshoot and, combined with the pre-calibrated transient current overshoot peak value and UVC power mapping relationship, calculates the current UVC absolute irradiance (uW / cm²). 2 ).

[0076] The system integrates the irradiance over time to calculate the cumulative bactericidal dose (mJ / cm). 2When the dosage reaches the standard for eliminating a specific virus (such as E. coli), disinfection is complete.

[0077] If [UVC overshoot == False] is detected, indicating no UVC overshoot, an alarm will be triggered to indicate that the light source is invalid and UVC sterilization has failed.

[0078] Example 2 Accordingly, this embodiment provides an ultraviolet detection method based on spatially hierarchical trap state modulation transient response, implemented using the system described in Embodiment 1 of this invention, including the following steps: Design a detection PIN photodiode with a multi-layer defect state having a spatial gradient along the incident light direction, and construct a differential hardware architecture based on the detection PIN photodiode to output a differential ultraviolet photocurrent signal; A high-frequency transient sampling window and a regular timing sampling window are set, and the time parameter for extracting features within the high-frequency transient sampling window is configured to map to the time constant of the physical structure of the PIN photodiode being detected; The acquisition of ultraviolet band features includes: extracting UVC, UVB, and UVA band features respectively in a high-frequency transient sampling window based on configured time parameters; and continuously sampling the signal at a conventional sampling rate lower than the high-frequency sampling rate within a conventional time-series sampling window to extract time-frequency domain features. Light source event identification is performed based on the ultraviolet band features and time-frequency domain features.

[0079] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, A and B simultaneously, or B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0080] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0081] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0082] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0083] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. An ultraviolet detection system based on spatially hierarchical trap state modulation transient response, characterized in that, include: A differential hardware architecture is used to output differential ultraviolet photocurrent signals. The differential hardware architecture is based on a detection PIN photodiode, which has a multi-layer defect state design with a spatial gradient along the incident light direction. This design is used to modulate the transient response time constants of photogenerated carriers in the UVC, UVB, and UVA bands at the physical layer, so that the dynamic characteristics of photogenerated carriers in different bands can be separated in a stepwise manner on the time axis. The feature extraction module performs intensive sampling of the differential ultraviolet photocurrent signal at a high-frequency sampling rate within the high-frequency transient sampling window that enters the transient feature extraction cycle, and extracts ultraviolet band features including UVC, UVB, and UVA band features; within the regular time-series sampling window after the high-frequency transient sampling window ends, the signal is continuously sampled at a regular sampling rate lower than the high-frequency sampling rate to extract time-frequency domain features. A multi-dimensional decision engine identifies light source events based on the ultraviolet band features and time-frequency domain features.

2. The ultraviolet detection system based on spatially hierarchical trap state modulation transient response according to claim 1, characterized in that, The differential hardware architecture is specifically as follows: The differential hardware architecture also includes a reference PIN photodiode, which is manufactured on the same substrate as the detection PIN photodiode and has a common spatial layered trap state design. The surface of the reference PIN photodiode is covered with an ultraviolet cut-off filter layer. The detection PIN photodiode is in an exposed state; The output terminals of the detection PIN photodiode and the reference PIN photodiode are connected to the differential front-end circuit. The differential operation cancels the visible light and near-infrared background photocurrent and outputs a differential ultraviolet photocurrent signal to the feature extraction module.

3. The ultraviolet detection system based on spatially hierarchical trap state modulation transient response according to claim 1, characterized in that, The detection PIN photodiode adopts a P++ / P+ / N- / N+ doping profile design with spatial gradient, and its specific structure includes: A shallow defect structure layer, corresponding to the physical absorption region of the UVC band, introduces a high concentration of surface defect states on the surface of the P++ doped material layer through a plasma treatment process to capture UVC photogenerated carriers with extremely shallow penetration depth. The intermediate defect structure layer corresponds to the physical absorption region of the UVB band. Medium-deep energy level traps are controllably introduced into the P+ doped material layer through a light mass ion implantation process to modulate the SRH recombination and release time of photogenerated carriers in the UVB band. The deep bulk region structure corresponds to the physical absorption region of the UVA band. The N-lightly doped layer and the N+ doped layer are intrinsic epitaxial defects. During detection, the N-lightly doped layer is not completely connected. UVA photogenerated carriers are absorbed in the deep undepleted bulk region of the N-lightly doped layer and the N+ doped layer and diffuse to the depletion region.

4. The ultraviolet detection system based on spatially hierarchical trap state modulation transient response according to claim 3, characterized in that, The time parameter for the feature extraction module to extract ultraviolet band features within the high-frequency transient sampling window is configured to map to the time constant of the detection PIN photodiode physical structure: The highest peak value of current overshoot in the 0 to 20 microsecond range is extracted and mapped to the UVC band characteristics dominated by the shallow defect structure layer. Extract the integral area of ​​the current curve bulge in the 50-100 microsecond interval to map the UVB band characteristics dominated by the middle defect structure layer. The average slope of the monotonically rising current in the range of 200 microseconds to 1000 microseconds is extracted to map the UVA band characteristics dominated by the deep body structure layer.

5. The ultraviolet detection system based on spatially hierarchical trap state modulation transient response according to claim 1, characterized in that, The extraction of time-frequency spectral features in the feature extraction module specifically involves: calculating the peak factor of the signal within the conventional time-series sampling window, and performing a fast Fourier transform to extract the periodic flicker features in the frequency domain to obtain a spectrum.

6. The ultraviolet detection system based on spatially hierarchical trap state modulation transient response according to claim 5, characterized in that, The multidimensional decision engine includes a logic for determining dangerous arc events: when the UVC band feature in the ultraviolet band is extracted in the high-frequency transient sampling window, and the peak factor calculated in the regular time-series sampling window is greater than or equal to a preset chaos threshold, it is determined that an arc discharge event exists in the environment.

7. The ultraviolet detection system based on spatially hierarchical trap state modulation transient response according to claim 5, characterized in that, The multidimensional decision engine also includes the judgment logic for AC-driven ultraviolet lamp events: when any one or more band features in the ultraviolet band features are extracted in the high-frequency transient sampling window, and the peak factor calculated in the conventional time-series sampling window is less than the preset chaos threshold, and the spectrum map has a main energy column and a decreasing harmonic column at the AC power grid fundamental frequency and its integer multiples, it is determined that there is an AC-driven ultraviolet light source event in the environment.

8. The ultraviolet detection system based on spatially hierarchical trap state modulation transient response according to claim 5, characterized in that, The multidimensional decision engine also includes the logic for calculating ultraviolet radiation power: when the multidimensional decision engine does not trigger the arc discharge event and the AC-driven ultraviolet lamp event, the system performs statistical averaging on the ultraviolet band features extracted by the feature extraction module, and inversely calculates the absolute radiation power values ​​of UVC, UVB, and UVA under the current environment based on the preset discrete mapping table or fitting curve.

9. The ultraviolet detection system based on spatially hierarchical trap state modulation transient response according to claim 1, characterized in that, The system also includes a drive and timing control module, which is configured to operate in a periodic pulse trigger mode to control the detection PIN photodiode to alternately be in a dormant release cycle and a transient feature extraction cycle on the time axis. During the sleep release cycle, the system is powered off to clear the accumulated charge of the multilayer defect states inside the detection PIN photodiode, forcing the trap energy level to return to the zero potential baseline.

10. An ultraviolet detection method based on spatially hierarchical trap state modulation transient response, characterized in that, The method is implemented based on the ultraviolet detection system based on spatially hierarchical trap state modulation transient response as described in any one of claims 1 to 9, and includes the following steps: Design a detection PIN photodiode with a multi-layer defect state having a spatial gradient along the incident light direction, and construct a differential hardware architecture based on the detection PIN photodiode to output a differential ultraviolet photocurrent signal; A high-frequency transient sampling window and a regular timing sampling window are set, and the time parameter for extracting features within the high-frequency transient sampling window is configured to map to the time constant of the physical structure of the PIN photodiode being detected; Based on the configured time parameters, the differential ultraviolet photocurrent signal is densely sampled at a high-frequency sampling rate within the high-frequency transient sampling window to extract ultraviolet band features including UVC, UVB, and UVA band features; within the regular time-series sampling window after the high-frequency transient sampling window ends, the signal is continuously sampled at a regular sampling rate lower than the high-frequency sampling rate to extract time-frequency domain features. Light source event identification is performed based on the ultraviolet band features and time-frequency domain features.

Images