Insulation discharge defect detection method and system based on solar blind ultraviolet and UVA dual-band

Abstract

The invention discloses an insulation discharge defect detection method and system based on solar-blind ultraviolet and UVA dual wavebands, and belongs to the technical field of electrical equipment insulation state detection.The method comprises the steps that a solar-blind ultraviolet waveband optical signal, a UVA ultraviolet waveband optical signal, a discharge current pulse signal, a solar-blind ultraviolet waveband image and a UVA waveband image of detected equipment are synchronously collected; carrying out lock-in amplification noise suppression on the UVA ultraviolet light signal, and extracting a photon event rate feature; extracting a light intensity peak value characteristic from the solar blind ultraviolet light signal, and calibrating a discharge quantity value through a nonlinear regression model in combination with a photon event rate; carrying out dynamic range expansion on the UVA ultraviolet image, and carrying out wavelet fusion on the UVA ultraviolet image and a solar blind ultraviolet image; fractal dimension features of the discharge channel are extracted from the fused image; and inputting the photon event rate, the light intensity peak value, the discharge capacity value and the fractal dimension features into a deep learning model, and outputting defect types and severity.

Description

Technical Field

[0001] The present invention belongs to the technical field of insulation status detection of power equipment, and specifically relates to a method and system for detecting insulation discharge defects based on dual-band solar-blind ultraviolet and UVA, which is used for quantitative analysis of discharge intensity and fault diagnosis of external insulation defects of insulators, bushings, etc. of high-voltage equipment. Background Art

[0002] With the widespread use of high-voltage power equipment, discharge defect detection in external insulation equipment (such as insulators and bushings) has become critical to ensuring the safe operation of power systems. Traditional detection methods rely primarily on ultraviolet imaging technology, which captures the ultraviolet light signals generated during the discharge process to identify defects. However, existing technologies have the following significant problems:

[0003] ① Single ultraviolet band (such as UVA or solar blind) detection is easily affected by environmental interference and cannot fully reflect the discharge characteristics.

[0004] ② Traditional ultraviolet imaging technology has a limited dynamic range, and the details of the discharge morphology are severely lost under weak light conditions.

[0005] ③ There is a lack of quantitative model for the correlation between discharge value and optical characteristics, and defect classification relies on manual experience, which is inefficient.

[0006] Prior art document 1 (CN104407277A) discloses a partial discharge monitoring device and method based on dual-band ultraviolet video multi-information fusion in the field of insulation status detection technology of power equipment. The dual-band ultraviolet binocular camera and ultraviolet sensor work together to collect discharge videos and quantitative signals of different ultraviolet bands, and use DSP multi-information fusion processing circuit to perform feature analysis, weight statistics and discharge positioning on the dual-channel video. However, it does not clearly distinguish the specific types of the two ultraviolet bands used, and there is a deficiency in that the band selection is not very targeted; it relies on weight statistical fusion and DSP processing, and there is a defect that the association between discharge value and defect type still requires manual experience judgment.

[0007] Prior art document 2 (CN116188505A) discloses an ultraviolet-infrared collaborative insulation equipment online monitoring system in the field of insulation status detection technology for power equipment. It uses an ultraviolet imager and an infrared camera to synchronously collect discharge spot and temperature rise data, and uses the SURF algorithm to achieve cross-spectral image registration. Combined with the RANSAC mismatch elimination technology, the abnormal temperature rise area and the discharge spot are fused into a single diagnostic image. However, it adopts ultraviolet and infrared collaborative detection, and the discharge spectrum information is not fully obtained; it relies on SURF+RANSAC registration and image superposition and fusion, and there is a defect that weak discharge morphology may not be clearly identified. Summary of the Invention

[0008] In order to address the deficiencies in the prior art, the present invention provides a method and system for detecting insulation discharge defects based on a dual-band spectrum of day-blind ultraviolet and UVA, which solves the problems of insufficient detection accuracy of a single band, limited dynamic range of weak-light imaging, and difficulty in quantitatively correlating discharge values ​​with defect types.

[0009] The present invention adopts the following technical solutions.

[0010] A first aspect of the present invention provides a method for detecting insulation discharge defects based on solar-blind ultraviolet and UVA dual-bands, the method comprising:

[0011] Synchronously collect the solar-blind ultraviolet band light signal, UVA ultraviolet band light signal, discharge current pulse signal, solar-blind ultraviolet band image and UVA band image of the device under test;

[0012] Perform phase-locked amplification noise suppression on UVA ultraviolet light signals and extract photon event rate characteristics;

[0013] The peak light intensity feature is extracted from the solar-blind ultraviolet light signal, and the discharge value is calibrated by a nonlinear regression model based on the photon event rate.

[0014] Expand the dynamic range of UVA images and perform wavelet fusion with solar-blind UV images;

[0015] Extract the fractal dimension features of discharge channels from fused images;

[0016] The photon event rate, light intensity peak, discharge value and fractal dimension features are input into the deep learning model to output the defect type and severity.

[0017] Optionally, performing phase-locked amplification noise suppression on the UVA ultraviolet light signal and extracting photon event rate characteristics includes:

[0018] Modulate the UVA signal at a set frequency;

[0019] Perform time window integration operation on the modulated signal and the reference signal;

[0020] Extract the denoised photon event signal based on the integration result;

[0021] Count the number of photon events per unit time and generate photon event rate characteristics.

[0022] Optionally, calibrating the discharge value by using a nonlinear regression model includes:

[0023] Establish a nonlinear regression model between photon event rate and peak light intensity;

[0024] Calibrate the model coefficients using a standard pulse discharge source;

[0025] Calculate the discharge value based on the model.

[0026] Optionally, the performing dynamic range expansion on the UVA image includes:

[0027] Perform histogram cropping and equalization on image blocks, and the cropping threshold is dynamically calculated based on the average value of the image grayscale histogram;

[0028] Filter denoising is performed using dual-parameter control of spatial proximity and grayscale similarity.

[0029] Optionally, performing wavelet fusion on the UVA image after dynamic range expansion and the solar-blind UV image includes:

[0030] Performing multi-layer wavelet decomposition on the solar-blind ultraviolet image and the UVA ultraviolet image after dynamic range expansion to obtain their respective corresponding low-frequency components and high-frequency components;

[0031] The low-frequency components are fused using a weighted fusion strategy based on the difference in signal-to-noise ratio between the solar-blind ultraviolet band and the UVA band;

[0032] For the high frequency components, the absolute value maximum method is used for fusion;

[0033] Performing inverse wavelet transform on the fused low-frequency components and the fused high-frequency components to reconstruct the fused image;

[0034] When the UVA band signal-to-noise ratio is lower than a set threshold, the weighted fusion strategy is automatically adjusted to increase the weight of the solar-blind ultraviolet band component.

[0035] Optionally, extracting the fractal dimension features of the discharge channel from the fused image includes:

[0036] A pixel-level vector positioning algorithm is used to determine the discharge channel contour;

[0037] The fractal dimension is calculated based on the minimum number of discharge channels covered by boxes of different sizes.

[0038] Optionally, after outputting the defect diagnosis result, the method further includes:

[0039] When the discharge value exceeds the set threshold, an early warning is triggered;

[0040] Predict insulation life based on the monthly average rate of change of discharge values;

[0041] Determining the diffusion risk of discharge channels based on the continuous change of fractal dimension.

[0042] A second aspect of the present invention provides an insulation discharge defect detection system based on solar-blind ultraviolet and UVA dual-bands, based on the insulation discharge defect detection method based on solar-blind ultraviolet and UVA dual-bands described in the first aspect of the present invention, the system comprises:

[0043] Dual-band UV sensor module, used to synchronously collect the solar-blind UV band light signal, UVA UV band light signal and discharge current pulse signal of the device under test;

[0044] A photoelectric synchronous analysis module, connected to the dual-band ultraviolet sensor module, is used to process the optical signal and calibrate the discharge value;

[0045] Dual-band UV imaging module, used to simultaneously capture images of the device under test in the solar-blind UV band and UVA band;

[0046] An image processing module, connected to the dual-band ultraviolet imaging module, for processing dual-band images and extracting features;

[0047] The defect diagnosis module is connected to the photoelectric synchronous analysis module and the image processing module, and is used to output the defect diagnosis result.

[0048] Optionally, the dual-band ultraviolet sensing module includes:

[0049] Solar-blind UV sensor, with a response wavelength of 240-280nm;

[0050] UVA ultraviolet sensor, response wavelength is 320-400nm;

[0051] Wideband current sensor with a bandwidth of 1Hz-10MHz.

[0052] Optionally, the image processing module includes:

[0053] Dynamic range extension unit, which processes UVA images using the CLAHE algorithm with adaptive histogram clipping threshold;

[0054] Wavelet fusion unit, which performs multi-scale decomposition and reconstruction on the solar-blind band image and the processed UVA image;

[0055] The fractal feature extraction unit calculates the fractal dimension of the discharge channel based on the box counting method.

[0056] Compared with the prior art, the beneficial effects of the present invention include at least:

[0057] Dual-band complementarity: Combining the anti-interference performance of the solar-blind band with the high sensitivity of the UVA band improves detection accuracy in complex environments.

[0058] Dynamic range improvement: Through UVA image enhancement and fusion algorithms, the detail retention rate of weak light discharge morphology is improved by more than 40%.

[0059] Quantitative diagnostic capability: Joint modeling of optical and electrical features enables contactless measurement of discharge values, with a defect classification accuracy of ≥95%. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a flow chart of a method provided according to an embodiment of the present invention;

[0061] Figure 2 is a system structure block diagram provided according to an embodiment of the present invention;

[0062] Figure 3 This is a flow chart of dual-band spectrum fusion and discharge value calibration provided in accordance with an embodiment of the present invention;

[0063] Figure 4 This is a deep learning model architecture diagram provided in accordance with an embodiment of the present invention. DETAILED DESCRIPTION

[0064] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0065] like Figure 1 As shown, the present invention provides a method for detecting insulation discharge defects based on dual-band solar-blind ultraviolet and UVA in Example 1, the method comprising the following steps:

[0066] Step 1: Synchronously collect the optical signals, images, and electrical signals of the device under test, including the optical signals, images, and discharge current pulse signals of the solar-blind ultraviolet band and the UVA ultraviolet band, and perform timing alignment and preprocessing.

[0067] Preferably, the step 1 comprises:

[0068] The sensor synchronously captures the solar blind and UVA (Ultraviolet A) discharge spectra and aligns the timing of the light and electrical signals;

[0069] The UVA signal is phase-locked and amplified to suppress ambient light noise.

[0070] Further preferably, the noise suppression of the UVA signal includes:

[0071] Adopting digital phase-locked amplification technology, the UVA signal is modulated into a square wave through a mechanical chopper (frequency 1kHz), and the ambient light noise is suppressed to a low-frequency component, and the signal-to-noise ratio is improved by ≥20dB;

[0072] The effective signal is extracted by digital phase-locked amplification technology. The formula is:

[0073]

[0074] Where S(t) is the UVA signal modulated by the mechanical chopper; sin(2πft) is the reference signal, whose frequency is synchronized with the mechanical chopper; T = 10 ms is the integration time window;

[0075] Extract the denoised UVA photon event signal for subsequent PHR statistics.

[0076] In step 2, the photon hit rate (PHR) is extracted based on the preprocessed UVA signal, the peak intensity of the solar blind band (PISB) is extracted from the solar blind signal, the dual-band photoelectric characteristics are jointly analyzed, and the discharge value is calibrated using a nonlinear regression model.

[0077] Preferably, if Figure 3 As shown, in step 2, calibrating the discharge value by using a nonlinear regression model includes:

[0078] The PHR in the UVA band and the PISB in the solar blind band were statistically analyzed to establish a nonlinear regression model with the discharge amount:

[0079] Q=k1·PHR+k2·PISB+γ·(PHR·PISB)

[0080] Among them, k1, k2, and γ are calibrated by a standard pulse discharge source, and the calibration error is ≤±3%.

[0081] Further preferably, the discharge value Q is calibrated by a nonlinear regression model, input into the defect diagnosis model and stored in the database, and the discharge threshold Q is set and regularly modified based on historical data. T , when Q>Q T When the threshold Q T Adjust accordingly based on the false alarm rate. For example, if the false alarm rate is > 10%, the discharge threshold is set to 1.2Q. T The adjustment range is adjusted according to the requirements for the false alarm rate. The false alarm rate calculation rules are as follows:

[0082]

[0083] Step 3: Dynamic range enhancement is performed on the UVA image, which is then fused with the solar-blind band image. The contour fractal features of the discharge channel are extracted based on the dual-band fused image.

[0084] Preferably, in step 3, performing dynamic range enhancement on the UVA image includes:

[0085] Traditional CLAHE uses a fixed threshold or global statistical value to crop the histogram, usually using larger blocks such as 64×64, which may cause the loss of local details and does not integrate a noise reduction module. This paper adopts an improved CLAHE algorithm, calculates an adaptive cropping threshold based on image characteristics, performs histogram cropping and equalization on UVA low-light image blocks (32×32), and adds bilateral filtering processing:

[0086]

[0087] Where H(i) is the number of pixels at the i-th gray level in the histogram; T clip is the clipping threshold, T clip = 0.02·average histogram value, the average histogram value is the average value of the number of all grayscale pixels; H clip (i) is the cropped histogram;

[0088] Combined with bilateral filtering for noise reduction, the signal-to-noise ratio (SNR) is increased to over 45dB. The kernel function is:

[0089]

[0090] Where I(x) and I(y) represent the grayscale values ​​at pixel positions x and y, respectively; the spatial standard deviation σ s =3 and gray value standard deviation σ r = 0.1 controls the spatial proximity and grayscale similarity of the filters respectively.

[0091] Compared with the traditional CLAHE algorithm, the improved CLAHE algorithm adopted in the present invention can achieve the following technical effects:

[0092] 1. The dynamic range of low-light images is expanded by more than 40%, making the originally invisible <5μW / cm 2 The morphology of weak discharge can be identified (dynamic threshold clipping);

[0093] 2. The 32×32 block strategy reduces the local contrast enhancement error to ≤±3%, reducing block artifacts by 50% compared to traditional methods;

[0094] 3. The SNR is increased to over 45dB (traditional CLAHE is about 32dB), effectively suppressing ambient light interference in the UVA band (bilateral filtering denoising processing).

[0095] Preferably, in step 3, the dual-band image fusion includes:

[0096] Combined with the solar-blind band image, the solar-blind and dynamic range-enhanced UVA image is subjected to a three-layer Daubechies-4 wavelet decomposition;

[0097] Weighted fusion of low-frequency coefficients (sun blind weight 0.7, UVA weight 0.3);

[0098] The solar-blind ultraviolet band (240-280nm) has minimal interference from sunlight, and its low-frequency component signal-to-noise ratio (SNR) is generally higher than that of the UVA band (320-400nm). Through experimental calibration, the solar-blind band's low-frequency coefficient weight is set to 0.7, which can retain more stable background structure and main discharge channel information; the UVA band weight is set to 0.3 to supplement low-light details while suppressing its low-frequency noise. In addition, when ambient lighting conditions change drastically (for example, rainy weather can cause increased noise in the UVA band), the system calculates the local contrast and SNR of the low-frequency components of the dual-band in real time and dynamically adjusts the weight ratio. When the UVA band SNR drops below the threshold (<40dB), the solar-blind weight is increased to 0.8 and the UVA weight is reduced to 0.2, prioritizing the structural integrity of the fused image.

[0099] The high frequency coefficient is maximized by taking the absolute value method.

[0100] After completing the wavelet coefficient processing, the fused image needs to be reconstructed through inverse wavelet transform. The specific steps are as follows:

[0101] The fused low-frequency and high-frequency coefficients are transformed using a three-layer inverse wavelet transform, reconstructing the high-frequency and low-frequency components at each scale layer by layer. The reconstructed components are then superimposed in the reverse order of the wavelet decomposition to generate a dual-band fused image. After reconstruction, the Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) are calculated between the fused image and the original image to ensure a reconstruction error of ≤2% and the absence of block artifacts or spectral distortion.

[0102] Preferably, in step 3, extracting the contour fractal features of the discharge channel includes:

[0103] The pixel-level vector positioning algorithm is used to extract the morphological features (fractal dimension) of the discharge channel contour based on the box counting method:

[0104]

[0105] The minimum number of boxes N(ε) covering the discharge channel varies with the box size ε, and the calculation error is ≤±0.02.

[0106] Step 4: Input the dual-band photoelectric features and image features into the pre-trained deep learning model to obtain the discharge defect type and confidence level of the device under test.

[0107] Preferably, step 4 includes:

[0108] The spectral features (PHR, PISB), image features (fractal dimension, area), and discharge value (Q) are input into the trained classification model to output the defect type and confidence level.

[0109] Preferably, if Figure 4 As shown, the deep learning model includes:

[0110] A database containing discharge spectra, image features, and defect types is constructed. The fused features are classified based on a deep learning model, and the defect types and severity are output. The defect types include surface creepage, tip corona, and spark discharge.

[0111] The deep learning model architecture is:

[0112] The backbone network uses ResNet-50 and introduces a channel attention mechanism (SEBlock) to enhance the feature weight of the discharge area;

[0113] The input features are spectral parameters (PHR, PISB), discharge value (Q), and image features (fractal dimension);

[0114] It uses the Softmax classifier and supports three types of defects and confidence output.

[0115] Preferably, the step 4 further comprises:

[0116] After obtaining the discharge defect type and confidence level of the equipment under test, store them in the database together with the discharge value Q, fractal dimension data, and time series data. The insulation life is predicted by statistically analyzing the monthly average change rate of the Q value. If the annual average growth rate of the Q value is >20%, it is recommended to shorten the maintenance cycle. If the Q value is stable but the fractal dimension continues to decline for three consecutive months, it indicates that there is a diffusion risk in the discharge channel and it needs to be checked in depth.

[0117] To address the problems of incomplete discharge spectrum information, low cross-spectral fusion efficiency and manual reliance on defect classification in existing methods, the present invention uses microsecond-level synchronous acquisition of optical and electrical signals, and calibrates the discharge value based on a nonlinear regression model of photon event rate (PHR) and peak light intensity. The calibration error is controlled within ±3%. Combining fractal dimension features with the ResNet-50+SEB deep learning model, automatic defect type classification is achieved with an accuracy rate of ≥95% and a confidence level of ≥98%, greatly improving diagnostic efficiency.

[0118] like Figure 2As shown, the present invention provides an insulation discharge defect detection system based on the dual bands of solar-blind ultraviolet and UVA in Example 2. Based on the insulation discharge defect detection method based on the dual bands of solar-blind ultraviolet and UVA described in Example 1, the system includes:

[0119] Dual-band UV sensing module, photoelectric synchronous analysis module, dual-band UV imaging module, image processing module and defect diagnosis module;

[0120] Dual-band UV sensor module, used to synchronously collect optical and electrical signals from the device under test, including optical signals in the solar-blind UV band and UVA UV band, as well as discharge current pulse signals;

[0121] A photoelectric synchronous analysis module, connected to the dual-band ultraviolet sensor module, is used to suppress noise and align timing of the optical signal, extract the photon event rate and light intensity peak characteristics, and calibrate the discharge value Q through a nonlinear regression model;

[0122] The dual-band UV imaging module is used to simultaneously capture images of the device under test in the day-blind UV band (240-280nm) and the UVA band (320-400nm).

[0123] An image processing module, connected to the dual-band ultraviolet imaging module, is used to perform dynamic range expansion, image fusion, and fractal feature extraction on the solar blind and UVA dual-band images to generate fused discharge morphology features;

[0124] The defect diagnosis module is connected to the photoelectric synchronous analysis module and the image processing module, and is used to input the photoelectric characteristics and image fractal characteristics into the deep learning model, where the photoelectric characteristics include PHR, PISB and Q, and output the defect type and severity.

[0125] Preferably, the dual-band ultraviolet sensor module includes:

[0126] Solar-blind UV sensor: Hamamatsu C13959-01 is selected, with a response wavelength of 240-280nm and a quantum efficiency of ≥30%, used to capture strong discharge signals and resist sunlight interference;

[0127] UVA sensor: Hamamatsu S1337, response wavelength 320-400nm, dark count rate ≤ 10cps, used to capture weak discharge photon events;

[0128] Wideband current sensor: equipped with Pearson4118, bandwidth 1Hz-10MHz, sensitivity 0.1V / A, synchronous measurement of discharge current pulses.

[0129] It should be noted that, in order to address the problems in the prior art such as the fuzzy selection of ultraviolet bands leading to weak anti-interference ability, the insufficient dynamic range under weak light conditions causing the loss of discharge morphology details, and the lack of quantitative correlation between discharge value and defect type relying on manual experience, the present invention provides a dual-band collaborative detection mechanism of day-blind ultraviolet and UVA, combined with the improved CLAHE (Contrast Limited Adaptive Histogram Equalization) algorithm and wavelet fusion technology, which achieves significant enhancement of anti-interference ability in complex environments and expansion of the dynamic range of weak light images by more than 40%, so that weak discharge morphology can be clearly identified, while the signal-to-noise ratio is improved to more than 45dB, effectively retaining the discharge details.

[0130] Preferably, the optoelectronic synchronization analysis module includes a GPS timing unit, which realizes microsecond-level synchronous acquisition of dual-band optical signals and electrical parameters, ensuring that the time domain alignment accuracy is ≤1μs; illustratively, the GPS timing unit can be selected from U-bloxNEO-M8N.

[0131] Preferably, the photoelectric synchronous analysis module further includes a digital phase-locked amplifier unit, which uses a mechanical chopper to modulate the UVA signal to suppress ambient light noise.

[0132] Preferably, the dual-band ultraviolet imaging module includes a solar-blind ultraviolet imaging unit and a UVA imaging unit, both of which adopt a parallel light path ultraviolet imaging method. The discharged ultraviolet light is filtered to suppress sunlight interference, focused to the ICCD photocathode window through the objective lens, and converted into a visible light image through the image intensifier. The difference between the solar-blind ultraviolet imaging unit and the UVA imaging unit lies in the different selection of filter lenses.

[0133] Preferably, the image processing module includes:

[0134] Dynamic range extension unit, using an improved CLAHE algorithm to enhance the contrast of UVA low-light images;

[0135] Wavelet fusion unit, which performs multi-scale decomposition and reconstruction of solar blind and UVA images;

[0136] The fractal feature extraction unit calculates the fractal dimension of the discharge channel based on the box counting method.

[0137] The present invention provides a method for detecting insulation discharge defects based on dual-band solar-blind ultraviolet and UVA in Example 2, based on the insulation discharge defect detection system based on dual-band solar-blind ultraviolet and UVA described in Example 1.

[0138] In Example 3, the present invention provides two application examples of a method and system for detecting external insulation discharge defects in the solar-blind ultraviolet and UVA dual-bands, including:

[0139] First application example: Detection of creepage defects on the bushing surface of a substation

[0140] Deploy dual-band sensors to simultaneously collect discharge spectra and current signals;

[0141] Analysis of the UVA band photon event rate (PHR = 1200cps) and the peak solar blinding intensity (PISB = 15μW / cm 2 ), calibrated discharge capacity Q = 2.3nC; fractal dimension (1.78) was extracted after image fusion;

[0142] The above data is input into the defect diagnosis model and is determined to be "surface creepage" with a confidence level of 98%.

[0143] Second application example: Online monitoring of insulator tip corona

[0144] Real-time acquisition of dual-band data detected a sudden increase in the UVA photon event rate (PHR increased from 500cps to 3000cps), and the discharge capacity was calibrated to Q = 5.1nC;

[0145] The discharge value Q is significantly higher than the threshold (4nC), the fusion image shows the fractal dimension of the tip discharge channel (1.35), and the model outputs a "tip corona" warning.

[0146] The present disclosure may be a system, method and / or computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0147] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A method for detecting insulation discharge defects based on dual-band solar-blind ultraviolet and UVA, characterized in that: The steps include: Synchronously collect the solar-blind ultraviolet band light signal, UVA ultraviolet band light signal, discharge current pulse signal, solar-blind ultraviolet band image and UVA band image of the device under test; Perform phase-locked amplification noise suppression on UVA ultraviolet light signals and extract photon event rate characteristics; The peak light intensity feature is extracted from the solar-blind ultraviolet light signal, and the discharge value is calibrated by a nonlinear regression model based on the photon event rate. Expand the dynamic range of UVA images and perform wavelet fusion with solar-blind UV images; Extract the fractal dimension features of discharge channels from fused images; The photon event rate, light intensity peak, discharge value and fractal dimension features are input into the deep learning model to output the defect type and severity.

2. The method for detecting insulation discharge defects based on dual-band solar-blind ultraviolet and UVA according to claim 1, characterized in that: The method of performing phase-locked amplification noise suppression on the UVA ultraviolet light signal and extracting the photon event rate feature includes: Modulate the UVA signal at a set frequency; Perform time window integration operation on the modulated signal and the reference signal; Extract the denoised photon event signal based on the integration result; Count the number of photon events per unit time and generate photon event rate characteristics.

3. The method for detecting insulation discharge defects based on solar-blind ultraviolet and UVA dual-bands according to claim 1, characterized in that: The method of calibrating the discharge value by using a nonlinear regression model includes: Establish a nonlinear regression model between photon event rate and peak light intensity; Calibrate the model coefficients using a standard pulse discharge source; Calculate the discharge value based on the model.

4. The method for detecting insulation discharge defects based on dual-band solar-blind ultraviolet and UVA according to claim 1, characterized in that: The dynamic range expansion of the UVA ultraviolet image includes: Perform histogram cropping and equalization on image blocks, and the cropping threshold is dynamically calculated based on the average value of the image grayscale histogram; Filter denoising is performed using dual-parameter control of spatial proximity and grayscale similarity.

5. The method for detecting insulation discharge defects based on solar-blind ultraviolet and UVA dual-bands according to claim 4, characterized in that: The wavelet fusion of the UVA image after dynamic range expansion and the solar-blind UV image comprises: Performing multi-layer wavelet decomposition on the solar-blind ultraviolet image and the UVA ultraviolet image after dynamic range expansion to obtain their respective corresponding low-frequency components and high-frequency components; The low-frequency components are fused using a weighted fusion strategy based on the difference in signal-to-noise ratio between the solar-blind ultraviolet band and the UVA band; For the high frequency components, the absolute value maximum method is used for fusion; Performing inverse wavelet transform on the fused low-frequency components and the fused high-frequency components to reconstruct the fused image; When the UVA band signal-to-noise ratio is lower than a set threshold, the weighted fusion strategy is automatically adjusted to increase the weight of the solar-blind ultraviolet band component.

6. The method for detecting insulation discharge defects based on solar-blind ultraviolet and UVA dual-bands according to claim 1, characterized in that: The fractal dimension feature of the discharge channel extracted from the fused image includes: A pixel-level vector positioning algorithm is used to determine the discharge channel contour; The fractal dimension is calculated based on the minimum number of discharge channels covered by boxes of different sizes.

7. The method for detecting insulation discharge defects based on solar-blind ultraviolet and UVA dual-bands according to claim 1, characterized in that: After outputting the defect diagnosis result, the method further includes: When the discharge value exceeds the set threshold, an early warning is triggered; Predict insulation life based on the monthly average rate of change of discharge values; Determining the diffusion risk of discharge channels based on the continuous change of fractal dimension.

8. An insulation discharge defect detection system based on solar-blind ultraviolet and UVA dual-bands, based on the insulation discharge defect detection method based on solar-blind ultraviolet and UVA dual-bands according to any one of claims 1 to 7, characterized in that: The system includes: Dual-band UV sensor module, used to synchronously collect the solar-blind UV band light signal, UVA UV band light signal and discharge current pulse signal of the device under test; A photoelectric synchronous analysis module, connected to the dual-band ultraviolet sensor module, is used to process the optical signal and calibrate the discharge value; Dual-band UV imaging module, used to simultaneously capture images of the device under test in the solar-blind UV band and UVA band; An image processing module, connected to the dual-band ultraviolet imaging module, for processing dual-band images and extracting features; The defect diagnosis module is connected to the photoelectric synchronous analysis module and the image processing module, and is used to output the defect diagnosis result.

9. The insulation discharge defect detection system based on solar-blind ultraviolet and UVA dual-band according to claim 8, characterized in that: The dual-band ultraviolet sensor module includes: Solar-blind UV sensor, with a response wavelength of 240-280nm; UVA ultraviolet sensor, response wavelength is 320-400nm; Wideband current sensor with a bandwidth of 1Hz-10MHz.

10. The insulation discharge defect detection system based on solar-blind ultraviolet and UVA dual-band according to claim 8, characterized in that: The image processing module includes: Dynamic range extension unit, which processes UVA images using the CLAHE algorithm with adaptive histogram clipping threshold; Wavelet fusion unit, which performs multi-scale decomposition and reconstruction on the solar-blind band image and the processed UVA image; The fractal feature extraction unit calculates the fractal dimension of the discharge channel based on the box counting method.

Images