Self-adaptive ultrahigh frequency partial discharge signal extraction device and method based on bimodal noise reference

Abstract

The invention discloses a self-adaptive ultrahigh frequency partial discharge signal extraction device and method based on dual-mode noise reference, and relates to the technical field of communication, and the device comprises a dual-sensor cooperative collection system which comprises a directional ultrahigh frequency sensor group and a noise feature learning module; the self-adaptive noise elimination device is connected with the dual-sensor collaborative acquisition system; and the intelligent signal identification system is connected with the self-adaptive noise elimination device. An auxiliary sensor in the directional ultrahigh frequency sensor group is specially used for collecting environmental background noise, and the noise feature learning module is used for extracting a time-frequency domain feature spectrum of the background noise by using Hilbert-Huang transform, so that the device can know the feature change of the noise in real time. Meanwhile, a dynamic threshold value adjusting unit in the self-adaptive noise elimination device can automatically adjust a signal triggering threshold value according to the noise energy level, automatic tracking of noise changes is achieved, and the adaptability of the device to the complex noise environment is greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of communication technologies, and particularly to an adaptive UHF partial discharge signal extraction device and method based on a dual-mode noise reference. Background Art

[0002] In the monitoring of UHF partial discharge signals in the field of communication, past technologies have faced many challenges. With the change of the environment, on-site UHF partial discharge signals are subject to varying degrees of noise interference, and the intensity and characteristics of the noise will also change accordingly. For example, in industrial sites, various types of electromagnetic noise are generated when electrical equipment operates, and these noises will be superimposed on the UHF partial discharge signals, making the monitored signals become complex.

[0003] Traditional on-line partial discharge monitoring technologies are difficult to automatically and accurately track noise changes. On the one hand, in the face of noises with different frequencies and intensities, there is a lack of an effective adaptive mechanism and it is unable to timely adjust the monitoring parameters to adapt to the changes in the noise environment. For example, when high-intensity pulse noise suddenly appears in the environment, traditional monitoring equipment may misjudge due to the inability to adjust the threshold in time, misidentifying the noise as a partial discharge signal. On the other hand, in terms of noise elimination, traditional technical means are limited. Existing filters can usually only handle specific types of noise, and it is difficult to achieve comprehensive and effective noise reduction for complex and variable noises. For example, although comb filters can suppress periodic interference, their processing effect on non-stationary signals and pulse-type burst noises is not good, which results in a large amount of noise still being included in the finally extracted partial discharge signals, affecting the accuracy and reliability of the signals.

[0004] In addition, in the differentiation between partial discharge signals and interference signals, traditional technologies lack an efficient discrimination method. Since there are certain similarities in the characteristics between typical partial discharge signals and industrial interference signals, relying solely on simple feature judgments is prone to misjudgment, thereby reducing the accuracy of the monitoring system. For example, the phase distribution maps and band energy ratios of some industrial interference signals are relatively close to those of partial discharge signals, and traditional methods are difficult to accurately distinguish between the two, resulting in deviations in the monitoring results. These problems seriously affect the accuracy and reliability of UHF partial discharge signal monitoring, making it difficult to achieve an ideal effect in the condition assessment of electrical equipment, and further affecting the safe and stable operation of the equipment. Summary of the Invention

[0005] To solve the above technical problems, an adaptive UHF partial discharge signal extraction device and method based on a dual-mode noise reference are provided, and this technical solution solves the above problems.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: An adaptive UHF partial discharge signal extraction device based on a dual-mode noise reference, comprising: Dual-sensor collaborative acquisition system, the dual-sensor collaborative acquisition system includes a directional ultra-high frequency sensor group and a noise feature learning module; Adaptive noise cancellation device, connected to the dual-sensor collaborative acquisition system; Intelligent signal discrimination system, connected to the adaptive noise cancellation device; Among them, the directional ultra-high frequency sensor group adopts an array antenna design, including a main sensor (UHF1) and an auxiliary sensor (UHF2). The main sensor (UHF1) is used to directionally receive the electromagnetic wave signal of the device body, and the auxiliary sensor (UHF2) is used to collect the environmental background noise; the noise feature learning module is used to extract the time-frequency domain feature map of the background noise through Hilbert-Huang transform.

[0007] Preferably, the adaptive noise cancellation device includes a dynamic threshold adjustment unit and a three-stage composite filter chain; The dynamic threshold adjustment unit automatically adjusts the signal trigger threshold based on the noise energy level, and the sensitivity range of the dynamic threshold adjustment unit is 0.01 - 100 mV; The three-stage composite filter chain is composed of a comb filter, a wavelet packet denoising module, and a neural network filter.

[0008] Preferably, the comb filter is used to suppress periodic interference, the wavelet packet denoising module is used to process non-stationary signals, and the neural network filter is used to identify pulse-type burst noise.

[0009] Preferably, the intelligent signal discrimination system includes a multi-dimensional feature library and an online pattern recognition engine; The multi-dimensional feature library contains the characteristic parameters of more than 50 typical partial discharge signals and 30 types of industrial interference signals; The online pattern recognition engine uses the DTW dynamic time warping algorithm to align the time-domain waveform and combines a convolutional neural network for discharge pattern classification.

[0010] Preferably, the specific steps of the Hilbert-Huang transform include: performing empirical mode decomposition on the collected background noise signal to obtain multiple intrinsic mode functions, and then performing Hilbert transform on each intrinsic mode function to obtain the time-frequency domain feature map of the background noise.

[0011] Preferably, the formula for the dynamic threshold adjustment unit to automatically adjust the signal trigger threshold is: T = k×E + b where T is the adjusted signal trigger threshold, E is the noise energy level, k and b are preset coefficients, and k and b are determined according to the actual application scenario.

[0012] Preferably, the formula for calculating the distance between two time-domain waveforms in the DTW (Dynamic Time Warping) algorithm is as follows: where x(i) and y(j) are the amplitudes of the two time-domain waveforms at times i and j respectively, and D(i, j) is the cumulative distance between the two waveforms at times i and j.

[0013] An adaptive UHF partial discharge signal extraction method based on a dual-modal noise reference includes: Directly receiving the electromagnetic wave signal of the equipment body through the main sensor (UHF1) of the directional UHF sensor group, and collecting the ambient background noise through the auxiliary sensor (UHF2); Using the noise feature learning module to extract the time-frequency domain feature map of the background noise through Hilbert-Huang transform; Transmitting the collected signal to the adaptive noise cancellation device. The adaptive noise cancellation device automatically adjusts the signal trigger threshold based on the noise energy level through the dynamic threshold adjustment unit, and filters the signal through a three-stage composite filter chain; Transmitting the filtered signal to the intelligent signal discrimination system. The intelligent signal discrimination system aligns the time-domain waveforms using the DTW (Dynamic Time Warping) algorithm through the online mode recognition engine, and classifies the discharge mode of the signal in combination with the convolutional neural network, so as to extract the UHF partial discharge signal.

[0014] Preferably, in the step of filtering the signal by the adaptive noise cancellation device, the comb filter first suppresses the periodic interference in the signal, then the non-stationary signal is processed by the wavelet packet denoising module, and finally the impulse burst noise is identified and eliminated by the neural network filter.

[0015] Preferably, the filtering formula of the comb filter is: In the formula, M is the filter order, and z is the complex variable; The wavelet packet denoising module adopts the wavelet packet decomposition formula, specifically: In the formula, h(n) and are the low-pass and high-pass filter coefficients respectively, j is the decomposition layer number, k is the node serial number, and t is the time variable.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: The auxiliary sensors in the directional ultra-high frequency sensor group are specifically used to collect ambient background noise, and the noise feature learning module extracts the time-frequency domain feature map of the background noise by using the Hilbert-Huang transform, which enables the device to understand the characteristic changes of the noise in real time. At the same time, the dynamic threshold adjustment unit in the adaptive noise cancellation device can automatically adjust the signal trigger threshold according to the noise energy level, realizing automatic tracking of noise changes and greatly improving the adaptability of the device to complex noise environments.

[0017] In terms of noise elimination, through the cooperation of the comb filter, wavelet packet denoising module and neural network filter, the periodic interference, non-stationary signals and impulse-like burst noises are processed respectively, comprehensively and effectively eliminating various noises, improving the purity of the signal, and making the extracted ultra-high frequency partial discharge signal more accurate and reliable.

[0018] Regarding the problem that it is difficult to distinguish between partial discharge signals and interference signals, by establishing a multi-dimensional feature library containing the characteristic parameters of various typical partial discharge signals and industrial interference signals, and using the DTW dynamic time warping algorithm combined with a convolutional neural network for discharge pattern classification, the discrimination ability of different signals is greatly improved, the false alarm rate is reduced, the accuracy of the monitoring results is ensured, providing a reliable basis for the condition assessment of electrical equipment, and ensuring the safe and stable operation of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is the composition diagram of the device of the present invention; Figure 2 is the flowchart of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0020] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations.

[0021] Referring to Figure 1 and Figure 2 as shown, the adaptive ultra-high frequency partial discharge signal extraction device based on a dual-modal noise reference includes: A dual-sensor collaborative acquisition system, which includes a directional ultra-high frequency sensor group and a noise feature learning module; An adaptive noise cancellation device, connected to the dual-sensor collaborative acquisition system; An intelligent signal discrimination system, connected to the adaptive noise cancellation device; Among them, the directional ultra-high frequency sensor group adopts an array antenna design, including a main sensor (UHF1) and an auxiliary sensor (UHF2). The main sensor (UHF1) is used to directionally receive the electromagnetic wave signal of the equipment body, and the auxiliary sensor (UHF2) is used to collect the environmental background noise. The noise feature learning module is used to extract the time-frequency domain feature map of the background noise through Hilbert-Huang transform.

[0022] Specifically, the device mainly consists of a dual-sensor collaborative acquisition system, an adaptive noise cancellation device, and an intelligent signal discrimination system. The directional ultra-high frequency sensor group in the dual-sensor collaborative acquisition system directionally receives the electromagnetic wave signal emitted by the equipment body through the main sensor (UHF1), which can accurately obtain the signal related to the partial discharge of the equipment. The auxiliary sensor (UHF2) is responsible for collecting the environmental background noise. The noise feature learning module uses Hilbert-Huang transform to process the background noise collected by the auxiliary sensor, decomposes the complex noise signal, and extracts the time-frequency domain feature map to analyze the characteristics of the noise. The adaptive noise cancellation device and the intelligent signal discrimination system respectively perform noise reduction and classification processing on the collected signals, and finally realize the extraction of the ultra-high frequency partial discharge signal. The dual-sensor collaborative acquisition system has a clear division of labor, making the acquisition of the equipment body signal and the background noise more targeted. Obtaining the noise time-frequency domain feature map through Hilbert-Huang transform provides an important basis for subsequent noise cancellation and signal processing, helps improve the adaptability of the entire device to the complex noise environment, and ensures the accuracy of the ultra-high frequency partial discharge signal extraction.

[0023] The adaptive noise cancellation device includes a dynamic threshold adjustment unit and a three-stage composite filter chain; The dynamic threshold adjustment unit automatically adjusts the signal trigger threshold based on the noise energy level, and the sensitivity range of the dynamic threshold adjustment unit is 0.01 - 100 mV; The three-stage composite filter chain consists of a comb filter, a wavelet packet denoising module, and a neural network filter.

[0024] Specifically, the dynamic threshold adjustment unit will automatically adjust the signal trigger threshold according to the change of the noise energy level, and its sensitivity can be adjusted between 0.01 - 100 mV. When the noise energy is high, the trigger threshold is appropriately increased to avoid misjudgment caused by noise interference. When the noise energy is low, the trigger threshold is decreased to ensure that weak partial discharge signals can also be captured. The three-stage composite filter chain consists of a comb filter, a wavelet packet denoising module, and a neural network filter, which processes different types of noise. The dynamic threshold adjustment unit improves the response ability of the device to noise changes and ensures the effectiveness of signal acquisition. The combination of the three-stage composite filter chain realizes the comprehensive suppression of various noises, greatly improves the signal quality, and makes the subsequent processing of the ultra-high frequency partial discharge signal more reliable.

[0025] The comb filter is used to suppress periodic interference, the wavelet packet denoising module is used to process non-stationary signals, and the neural network filter is used to identify impulse-type burst noise.

[0026] Specifically, the comb filter utilizes its own filtering characteristics to suppress periodic interference in the signal. It can identify and remove noise signals with periodic patterns. The wavelet packet denoising module is good at processing non-stationary signals, whose statistical characteristics change over time. The wavelet packet denoising module decomposes the signal into multiple layers and decomposes the signal into different frequency bands, thereby effectively removing non-stationary noise. The neural network filter uses the powerful pattern recognition ability of the neural network to identify and eliminate impulse-type burst noise. The three filters each perform their own functions and work together to comprehensively eliminate noises with different characteristics, making the filtered signal purer, improving the extraction accuracy of the UHF partial discharge signal, and reducing the interference of noise on signal analysis.

[0027] The intelligent signal discrimination system includes a multi-dimensional feature library and an online pattern recognition engine; The multi-dimensional feature library contains characteristic parameters of more than 50 typical partial discharge signals and 30 types of industrial interference signals; The online pattern recognition engine uses the DTW dynamic time warping algorithm to align the time-domain waveforms and combines a convolutional neural network for discharge pattern classification.

[0028] Specifically, the intelligent signal discrimination system consists of a multi-dimensional feature library and an online pattern recognition engine. The multi-dimensional feature library stores characteristic parameters of more than 50 typical partial discharge signals and 30 types of industrial interference signals, and these parameters cover multiple aspects such as the phase distribution map, pulse rise time, and frequency band energy ratio of the signal. The online pattern recognition engine uses the DTW dynamic time warping algorithm to align the collected time-domain waveforms with the standard waveforms in the feature library, solving the problem that it is difficult to compare due to inconsistent signal time scales. Then, combining the powerful image recognition and classification ability of the convolutional neural network, the discharge pattern of the signal is classified to determine whether the signal is a partial discharge signal or an interference signal. The multi-dimensional feature library provides a rich reference basis for signal classification. The combination of the DTW dynamic time warping algorithm and the convolutional neural network greatly improves the accuracy of signal discrimination, effectively avoids misjudging interference signals as partial discharge signals, and improves the reliability of UHF partial discharge signal extraction.

[0029] The specific steps of the Hilbert-Huang transform include: performing empirical mode decomposition on the collected background noise signal to obtain multiple intrinsic mode functions, and then performing Hilbert transform on each intrinsic mode function to obtain the time-frequency domain characteristic map of the background noise.

[0030] Specifically, the Hilbert-Huang transform first performs empirical mode decomposition on the collected background noise signal. Empirical mode decomposition decomposes a complex signal into several intrinsic mode functions, which represent the characteristics of the signal at different time scales. Then, the Hilbert transform is performed on each intrinsic mode function to transform the time-domain signal into the frequency domain, and finally, the time-frequency domain characteristic spectrum of the background noise is obtained, thus clearly showing the distribution characteristics of the noise in time and frequency. This method can deeply analyze the characteristics of the background noise, provide detailed noise information for the adaptive noise cancellation device and the intelligent signal discrimination system, help the device better adapt to the noise environment, and improve the extraction and analysis capabilities of the ultra-high frequency partial discharge signal.

[0031] The formula for the dynamic threshold adjustment unit to automatically adjust the signal trigger threshold is: T = k × E + b Where T is the adjusted signal trigger threshold, E is the noise energy level, and k and b are preset coefficients, and k and b are determined according to the actual application scenario.

[0032] Specifically, through this automatic adjustment mechanism, the device can flexibly adapt to environments with different noise intensities, avoid false triggering in high-noise environments, and not miss weak partial discharge signals in low-noise environments, improving the accuracy and reliability of signal acquisition.

[0033] The formula for calculating the distance between two time-domain waveforms in the DTW dynamic time warping algorithm is: Where x(i) and y(j) are the amplitudes of the two time-domain waveforms at times i and j respectively, and D(i,j) is the cumulative distance between the two waveforms at times i and j.

[0034] Specifically, the algorithm finds the optimal time warping path, enabling the two waveforms to be better aligned in time, facilitating subsequent comparison and classification. It solves the problem of inconsistent time scales of different time-domain waveforms, allowing the collected signals to accurately match the standard waveforms in the feature library, providing a more reliable data basis for the convolutional neural network to classify discharge patterns, and improving the accuracy of signal classification.

[0035] The adaptive ultra-high frequency partial discharge signal extraction method based on dual-mode noise reference includes: Directly receiving the electromagnetic wave signal of the device body through the main sensor (UHF1) of the directional ultra-high frequency sensor group, and the auxiliary sensor (UHF2) collects the environmental background noise; Using the noise feature learning module to extract the time-frequency domain characteristic spectrum of the background noise through the Hilbert-Huang transform; The collected signals are transmitted to an adaptive noise cancellation device. The adaptive noise cancellation device automatically adjusts the signal trigger threshold based on the noise energy level through a dynamic threshold adjustment unit, and filters the signals through a three-stage composite filter chain. The filtered signals are transmitted to an intelligent signal discrimination system. The intelligent signal discrimination system uses the DTW (Dynamic Time Warping) algorithm to align the time-domain waveforms through an online mode recognition engine, and classifies the signals in combination with a convolutional neural network to extract the ultra-high frequency partial discharge signals.

[0036] Specifically, first, the main and auxiliary sensors of the directional ultra-high frequency sensor group are used to collect the electromagnetic wave signals of the equipment body and the ambient background noise respectively. Then, the noise feature learning module performs Hilbert-Huang transform on the background noise to obtain its characteristic spectrogram. The collected signals are transmitted to the adaptive noise cancellation device and are filtered through dynamic threshold adjustment and a three-stage composite filter chain. Finally, the filtered signals enter the intelligent signal discrimination system, and the discharge mode classification is carried out through the DTW algorithm and the convolutional neural network to extract the ultra-high frequency partial discharge signals. Each step of this method is closely linked, forming a complete ultra-high frequency partial discharge signal extraction process from signal collection, noise analysis, noise reduction processing to signal classification, ensuring the efficiency and accuracy of signal extraction.

[0037] In the step of filtering the signals by the adaptive noise cancellation device, the comb filter first suppresses the periodic interference in the signals, then the non-stationary signals are processed by the wavelet packet denoising module, and finally the impulse-like burst noise is identified and eliminated by the neural network filter.

[0038] The filtering formula of the comb filter is: In the formula, M is the filter order, and z is the complex variable; The wavelet packet denoising module adopts the wavelet packet decomposition formula, specifically: In the formula, h(n) and are the low-pass and high-pass filter coefficients respectively, j is the decomposition layer number, k is the node serial number, and t is the time variable.

[0039] Specifically, when the adaptive noise cancellation device filters a signal, the comb filter first suppresses the periodic interference in the signal and removes the periodic noise components. Then, the wavelet packet denoising module further processes the signal processed by the comb filter. In view of the characteristics of non-stationary signals, the signal is decomposed into different frequency bands for denoising. Finally, the neural network filter identifies and eliminates the possible impulse-like burst noise in the signal. At the same time, these formulas provide a theoretical basis for the design and optimization of the filters, enabling the comb filter and the wavelet packet denoising module to process the corresponding noise more accurately, improving the performance of the entire noise cancellation device, and helping to extract the ultra-high frequency partial discharge signal more accurately.

[0040] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification is only the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and all these changes and improvements fall within the scope of the present invention claimed.

Claims

1. An adaptive UHF partial discharge signal extraction device based on a bimodal noise reference, characterized in that Including: A dual-sensor collaborative acquisition system, which includes a directional ultra-high frequency sensor group and a noise feature learning module; An adaptive noise cancellation device, connected to the dual-sensor collaborative acquisition system; An intelligent signal discrimination system, connected to the adaptive noise cancellation device; Among them, the directional ultra-high frequency sensor group adopts an array antenna design, including a main sensor (UHF1) and an auxiliary sensor (UHF2). The main sensor (UHF1) is used to directionally receive the electromagnetic wave signal of the equipment body, and the auxiliary sensor (UHF2) is used to collect the environmental background noise; the noise feature learning module is used to extract the time-frequency domain feature map of the background noise through Hilbert-Huang transform.

2. The adaptive UHF partial discharge signal extraction device based on a dual-modal noise reference according to claim 1, wherein The adaptive noise cancellation device includes a dynamic threshold adjustment unit and a three-stage composite filter chain; The dynamic threshold adjustment unit automatically adjusts the signal trigger threshold based on the noise energy level, and the sensitivity range of the dynamic threshold adjustment unit is 0.01-100 mV; The three-stage composite filter chain is composed of a comb filter, a wavelet packet denoising module, and a neural network filter.

3. The adaptive UHF partial discharge signal extraction device based on a dual-modal noise reference according to claim 1, wherein The comb filter is used to suppress periodic interference, the wavelet packet denoising module is used to process non-stationary signals, and the neural network filter is used to identify pulse-type burst noise.

4. The adaptive UHF partial discharge signal extraction device based on a bimodal noise reference according to claim 1, characterized in that, The intelligent signal discrimination system includes a multi-dimensional feature library and an online pattern recognition engine; The multi-dimensional feature library contains the characteristic parameters of more than 50 typical partial discharge signals and 30 types of industrial interference signals; The online pattern recognition engine uses the DTW dynamic time warping algorithm to align the time-domain waveforms and combines a convolutional neural network for discharge pattern classification.

5. The adaptive UHF partial discharge signal extraction device based on dual-mode noise reference according to claim 1, characterized in that The specific steps of the Hilbert-Huang transform include: performing empirical mode decomposition on the collected background noise signal to obtain multiple intrinsic mode functions, and then performing Hilbert transform on each intrinsic mode function to obtain the time-frequency domain feature map of the background noise.

6. The adaptive UHF partial discharge signal extraction device based on a dual-modal noise reference according to claim 1, wherein The formula for the dynamic threshold adjustment unit to automatically adjust the signal trigger threshold is: T = k×E + b Where, T is the adjusted signal trigger threshold, E is the noise energy level, k and b are preset coefficients, and k and b are determined according to the actual application scenario.

7. The adaptive UHF partial discharge signal extraction device based on dual-mode noise reference according to claim 1, wherein The formula for calculating the distance between two time-domain waveforms in the DTW dynamic time warping algorithm is: Where, x(i) and y(j)) are the amplitudes of the two time-domain waveforms at times i and j respectively, and D(i,j) is the cumulative distance between the two waveforms at times i and j.

8. An adaptive UHF partial discharge signal extraction method based on a bimodal noise reference, and an adaptive UHF partial discharge signal extraction device based on the bimodal noise reference according to any one of claims 1-7, characterized in that, Including: Directionally receive the electromagnetic wave signal of the equipment body through the main sensor (UHF1) of the directional ultra-high frequency sensor group, and the auxiliary sensor (UHF2) collects the environmental background noise; Use the noise feature learning module to extract the time-frequency domain feature map of the background noise through Hilbert-Huang transform; Transmit the collected signal to the adaptive noise cancellation device, and the adaptive noise cancellation device automatically adjusts the signal trigger threshold based on the noise energy level through the dynamic threshold adjustment unit, and filters the signal through the three-stage composite filter chain; Transmit the filtered signal to the intelligent signal discrimination system. The intelligent signal discrimination system uses the DTW (Dynamic Time Warping) algorithm through an online mode recognition engine to align the time-domain waveforms, and combines a convolutional neural network to classify the signal discharge mode, thereby extracting the UHF partial discharge signal.

9. The adaptive UHF partial discharge signal extraction device and method based on dual-modal noise reference according to claim 1, characterized in that, In the step of the adaptive noise cancellation device filtering the signal, the comb filter first suppresses the periodic interference in the signal, then the non-stationary signal is processed by the wavelet packet denoising module, and finally the neural network filter identifies and eliminates the impulse-type burst noise.

10. The adaptive UHF partial discharge signal extraction device and method based on dual-mode noise reference according to claim 1, characterized in that, The filtering formula of the comb filter is: Where M is the filter order and z is a complex variable; The wavelet packet denoising module uses the wavelet packet decomposition formula, specifically: where h(n) and are the low-pass and high-pass filter coefficients respectively, j is the decomposition level, k is the node number, and t is the time variable.

Images