Pulse laser echo signal enhancement method based on digital signal processing

Abstract

The invention discloses a pulse laser echo signal enhancement method based on digital signal processing, and provides a target waveform adaptive adjustment module based on a block function chain type adaptive filter in combination with a multi-pulse superposition and matched filtering technology, so that Gaussian and non-Gaussian noise causing waveform distortion can be filtered out, and the signal-to-noise ratio is effectively improved. Based on the FPGA parallel processing capability, a data expansion module is adopted, weight updating is accelerated under the same learning step length, and the problems that laser echo signals in a scattering medium fluctuate obviously and the algorithm instantaneity is insufficient are solved. A data screening module is adopted to preprocess superposed signals, non-Gaussian background noise and peak noise are filtered out, and high-quality expected data are provided for a matched filter. According to the invention, when the echo signal of the laser detection system is a weak signal, the signal-to-noise ratio of the target wave crest is effectively improved, and meanwhile, the real-time processing capability of the algorithm is greatly improved.

Description

Technical Field

[0001] This invention belongs to the field of laser detection, specifically relating to a method for enhancing pulsed laser echo signals based on digital signal processing. Background Technology

[0002] Semiconductor lasers possess advantages such as high repetition rate, low power consumption, small size, and light weight, making them widely used in laser ranging and radar systems. However, laser detection suffers from backscattering interference in aerosol and aquatic environments, thus limiting its emission power. Conversely, low laser emission power leads to weak echo signals that are difficult to detect.

[0003] Currently, when the laser echo signal is weak, the main problem is how to improve the signal-to-noise ratio to separate the target echo from the noise signal. In the laser echo signal, the noise is mainly Gaussian noise, that is, random noise signal that conforms to the Gaussian distribution law (Hu Wei, Sun Xiaoquan, Dou Xianan. Analysis of the influence of sampling clock jitter in weak light signal detection [J]. Laser & Infrared, 2014, 44(07): 824-828.). Therefore, scholars mainly use the method of multi-pulse superposition to filter out the Gaussian noise. Zhang Yu et al. of Harbin Institute of Technology conducted computer simulation and laboratory experiments. The experimental data showed that multi-pulse superposition laser multi-pulse accumulation detection can effectively improve the signal-to-noise ratio of laser detection. The application of accumulation detection can effectively detect the weak laser signal from the strong noise background, proving the correctness of the theory in engineering practice (Zhang Yu, Zhao Yuan, Yang Yong, et al. Computer simulation and experimental study of laser multi-pulse detection [J]. Infrared & Laser Engineering, 2007, (01): 60-63.). Liu Huihui et al. from Beijing Institute of Petrochemical Technology adopted a multi-pulse cross-correlation accumulation method, utilizing the uncorrelation of noise and using a DSP to perform a large number of addition operations, thereby achieving signal enhancement. In clear weather, the measurement distance was increased from less than 300m to more than 1000m (Liu Huihui, Zhang Ze, Liang Tiantai. Application of multi-pulse cross-correlation detection method in long-distance pulse laser ranging [J]. New Technology and New Process, 2015, (07): 140-142.). Liu Huihui et al. further improved the detection range by upgrading the hardware and using a DSP to meet the large computational requirements, but the DSP's processing speed is at the millisecond level, which is insufficient for real-time processing of laser detection. Zhang Yue from Changchun University of Science and Technology proposed a wavelet threshold denoising method based on EMD. This method employs non-uniformly spaced multi-pulse superposition, and uses the EMD algorithm to decompose the echo signal into different components according to frequency from high to low. Wavelet threshold denoising is then applied to the high-frequency components, increasing the laser echo signal-to-noise ratio from -15dB to 2.66dB. Using a 200Msps ADC for sampling, the target positioning error can be controlled within 1.5m. (Zhang Yue, Wang Chunyang, Liu Yanyan, et al. Research on Echo Processing Method of Multi-Pulse Remote Laser Ranging [J]. Information and Communication, 2018, (01): 48-50.)

[0004] Currently, researchers have made some progress in multi-pulse superposition methods to address the challenge of weak echo signals. However, these methods do not consider the potential broadening, shifting, and fluctuations in the laser echo waveform caused by strong scattering environments such as aerosols and water bodies. Superposition in these environments can lead to echo peak shifts, affecting detection performance. Furthermore, the computational demands of superposition algorithms increase the hardware requirements for ensuring real-time performance of the detection system. Therefore, this invention proposes a pulsed laser echo signal enhancement method based on digital signal processing to solve these problems. Summary of the Invention

[0005] The purpose of this invention is to provide a pulsed laser echo signal enhancement method based on digital signal processing. Combining multi-pulse superposition technology and matched filtering technology, a target waveform adaptive adjustment module based on Block Functional Link Adaptive Filter (BFLAF) is proposed to effectively improve the target peak signal-to-noise ratio. Based on the parallel processing capability of FPGA, a data expansion module and a data filtering module are adopted to greatly improve the real-time processing capability of the algorithm.

[0006] The technical solution for achieving the present invention is as follows: a method for enhancing pulsed laser echo signals based on digital signal processing, comprising the following steps:

[0007] Step 1: Obtain the laser echo signal based on full waveform sampling:

[0008] The laser emitter directs a collimated beam onto an underwater target, while the laser receiver captures all diffusely reflected photons within its field of view, acquiring the laser echo signal based on full-waveform sampling. Proceed to step 2.

[0009] Step 2: Combine the three consecutive laser echo signals acquired as a group and perform multi-pulse superposition to obtain the superimposed signal. Proceed to step 3.

[0010] Step 3: Design a block function chain adaptive filter, using the laser emission pulse waveform as the initial old signal template. Proceed to step 4.

[0011] Step 4: Replace the old signal template Input extended data block to obtain extended matrix The nonlinear extended matrix is ​​obtained by extending it with a nonlinear function. Combined with block filter weights After parallel-to-serial conversion, a new signal template for the matched filter is obtained. Finally, the new signal template The signal obtained after superposition of multiple pulses Perform matched filtering to obtain the enhanced output signal. Proceed to step 5.

[0012] Step 5: Combine the signals after multi-pulse superposition Import data filtering module, from Extract valid data As expected data for block function chain-type adaptive filters, to update the block filter weights. The updated result For the next round of block filter weights , will the new signal template As the old signal template for the next round Return to step 4.

[0013] Compared with the prior art, the significant advantages of this invention are:

[0014] (1) This invention combines multi-pulse superposition and matched filtering techniques to propose a target waveform adaptive adjustment module based on a Block Functional Link Adaptive Filter (BFLAF). Through multi-pulse superposition, the signal-to-noise ratio of the echo signal can be improved after m pulse superpositions. This effectively filters out Gaussian noise. Simultaneously, combined with matched filtering technology, the echo signal undergoes waveform adjustment and matching, filtering out non-Gaussian noise data that causes waveform distortion, further improving the signal-to-noise ratio.

[0015] (2) This invention proposes a hardware implementation method based on a block function chain adaptive filter, combined with the BFLAF algorithm. The FLAF algorithm is improved based on FPGA. By utilizing the parallel processing capability of FPGA and adopting a data expansion module, a block function chain adaptive filter is proposed, which improves the running speed of the FLAF algorithm. Under the same learning step size, the weight update speed is increased by L times, which solves the problems of significant fluctuation of laser detection target signal in scattering medium and insufficient real-time performance of algorithm during matched filtering.

[0016] (3) The present invention employs a data filtering module for filtering the superimposed signals. Preprocessing is performed to filter out non-Gaussian background noise or spike noise in advance, thereby obtaining effective data. The matched filter is used to update the desired data of the signal template, providing a higher quality signal for subsequent data processing. Simultaneously, the data extension length is determined based on the number of valid data points. When both are equal, the target waveform adaptive adjustment module only needs to run once to update the target waveform. Therefore, until the next multi-pulse superposition... Before its arrival, the FPGA had ample runtime to process the nonlinear function extension and weight update parts of the BFLAF algorithm, greatly improving the algorithm's real-time processing capabilities. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of an underwater laser detection system.

[0018] Figure 2 This is a schematic diagram illustrating the principle of multi-pulse superposition.

[0019] Figure 3 The schematic diagram of the block function chain-type adaptive filter designed for this invention.

[0020] Figure 4 This is a schematic diagram of the data filtering module of the present invention.

[0021] Figure 5 This is a schematic diagram of the single-element nonlinear function expansion principle in the data expansion block of this invention.

[0022] Figure 6 The schematic diagram of the data extension block designed for this invention.

[0023] Figure 7 This is a normalized diagram of the laser emission signal.

[0024] Figure 8 Comparison of weak echo signals before and after filtering using the method of this invention.

[0025] Figure 9 Comparison of strong echo signals before and after filtering using the method of this invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be fully and clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments and do not include all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0027] The following section will further introduce the specific implementation method, as well as the technical difficulties and inventive points of this invention, using this design example as an example.

[0028] Combination Figures 1-6 A method for enhancing pulsed laser echo signals based on digital signal processing, the specific steps of which are as follows:

[0029] This embodiment uses an underwater laser detection system as an example. If it is to be applied to other detection systems, only the corresponding number of superpositions and the original signal template need to be changed.

[0030] Step 1: Obtain the laser echo signal based on full waveform sampling:

[0031] Combination Figure 1 An underwater laser detection system is constructed, comprising a laser emitter, a laser receiver, and an underwater target. The laser emitter directs a collimated beam onto the target, and the laser receiver receives diffusely reflected photons within the field of view. The laser echo signal is acquired based on full waveform sampling. The signal is then transmitted to the FPGA via a data interface for further digital signal processing, proceeding to step 2.

[0032] Step 2: Combine the three consecutive laser echo signals acquired as a group and perform multi-pulse superposition to obtain the superimposed signal. The details are as follows:

[0033] Since the laser echo signal becomes a discrete signal after being sampled by the ADC, all signals are discrete sequences during data processing.

[0034] in, This represents the signal after superposition in group a. Whenever a new laser echo signal is acquired, the earliest acquired laser echo signal is discarded, ensuring that each group has exactly three laser echo signals. For example, Depend on It is obtained by superimposing a set of data. Depend on As a set of superimposed results... and so on.

[0035] Combination Figure 2 Since the emission frequency of the pulsed laser is 10kHz, which is much higher than the speed of the target in the water, the time interval between the emission of two pulsed laser beams is... Within this range, the change in the target's attitude is negligible. Therefore, it can be approximated that the laser receiver receives several consecutive segments of laser echo signals ①②③ (i.e., ... The results are the same. Based on this principle, multiple consecutive laser echo signals can be acquired within the effective acquisition time. The signals are superimposed on each other, and after three superpositions, the amplitude increases from V to 3V, and the signal-to-noise ratio is significantly improved. This is because the laser echo signal consists of periodic target echoes and random noise. After multi-pulse superposition, the amplitude of the target echoes is superimposed, while the random noise cancels each other out during the superposition, thereby improving the signal-to-noise ratio of the laser echo signal.

[0036] Laser echo signal acquired based on full waveform sampling Make the target echo as Random noise is Then we have:

[0037] (1),

[0038] Subscript Indicates the corresponding segment number.

[0039] The echo signal after m multi-pulse superposition for:

[0040] (2),

[0041] In equation (2), Indicates the first Segment laser echo signal, Indicates the first Target echo in segment laser echo signal, Indicates the first Random noise in segment laser echo signals.

[0042] Due to the target echo Since it is a periodic signal, therefore:

[0043] (3),

[0044] Due to Gaussian noise It is random, and after multiple pulses are superimposed, they will cancel each other out. Therefore, the effective value of the superimposed noise is... for:

[0045] (4),

[0046] In the formula, Random noise in the first segment of the laser echo signal of group a The effective value of represents the random noise before superposition. Therefore, the signal-to-noise ratio improvement after superposition is expressed as... express:

[0047] (5),

[0048] From equation (5), it can be seen that after m superpositions, the signal-to-noise ratio of the echo signal can be improved. Theoretically, the more times the signal-to-noise ratio (SNR) is superimposed, the greater the improvement. However, multi-pulse superposition essentially improves the SNR by reducing the detection frequency of the detection system. Considering the stability and dynamic performance of the detection system, the superposition count is set to 3. The corresponding set of laser echo signals acquired by the data interface are superimposed. Taking three consecutive laser emission pulses as a group, each pulse produces an echo signal after illuminating the target. The corresponding set of echo signals acquired by the data interface are then superimposed.

[0049] In summary, the combination of multi-pulse superposition and matched filtering techniques proposed in this invention can improve the signal-to-noise ratio of the echo signal after m pulse superpositions. times.

[0050] Proceed to step 3.

[0051] Step 3: Design a block function chain adaptive filter, using the laser emission pulse waveform as the initial old signal template. Proceed to step 4.

[0052] Step 4: Replace the old signal template (in Indicates the number of iterations. (Initially 0) Input extended data block to obtain extended matrix The nonlinear extended matrix is ​​obtained by extending it with a nonlinear function. Combined with block filter weights After parallel-to-serial conversion, a new signal template for the matched filter is obtained. Finally, the new signal template... and obtained after multi-pulse superposition Perform matched filtering to obtain the enhanced output signal. The details are as follows:

[0053] Combination Figures 3-4 The laser emission pulse waveform is used as the initial old signal template. and import extended data blocks. middle. For one OK, The data matrix of columns, where To extend the length of the data, This represents the number of filter taps.

[0054] Let the row vector of the data block be , ,but

[0055] (6),

[0056] Where T represents transpose. express The Middle Each element.

[0057] Combination Figure 5 The block function chain adaptive filter designed in this invention adopts a nonlinear function extension module based on trigonometric functions and is implemented in combination with the BFLAF algorithm.

[0058] Because the nonlinear extension of trigonometric functions has the advantages of strong nonlinear approximation ability and low computational complexity, its function expression is: ,in Let be the number of filter taps in the extension module, and also the number of functions in the extension module, then the filter's tap number is... ( The nonlinear extension expression for ) taps is:

[0059] (7),

[0060] in, express The Middle One element, , To expand the index, For extended order, This represents the length of the discrete sequence.

[0061] Since the number of functions in the extension module is Then the extended order for:

[0062] (8),

[0063] Let the old signal template be input After expansion using a nonlinear function, the column expansion matrix is ​​obtained. :

[0064] (9),

[0065] The obtained extended data matrix is ​​then extended using a nonlinear function to obtain a nonlinear extended matrix. for:

[0066] (10),

[0067] Then the new signal template of the matched filter for:

[0068] (11),

[0069] In the formula, the block filter weights This also represents the weight vector of the linear filtering part of the BFLAF algorithm.

[0070] The obtained new signal template of the matched filter and the signal after multi-pulse superposition After matched filtering, the enhanced output signal is obtained. , can be represented as:

[0071] (12),

[0072] In the formula, n represents the length of the discrete sequence, and s represents the discrete integral variable.

[0073] The innovation of this invention lies in proposing a target waveform adaptive adjustment module based on a block function chain adaptive filter, implemented in conjunction with the BFLAF algorithm. Based on FPGA, the traditional FLAF algorithm is improved. Utilizing the parallel processing capabilities of the FPGA, a block function chain adaptive filter is proposed, which enhances the running speed of the traditional FLAF algorithm. This achieves an L-fold increase in weight update speed while maintaining the same learning step size, solving the problems of significant fluctuations in underwater laser detection target signals and insufficient real-time performance during matched filtering.

[0074] Proceed to step 5.

[0075] Step 5: Combine the signals after multi-pulse superposition Import data filtering module, from Extract valid data As expected data for block function chain-type adaptive filters, to update the block filter weights. The updated result For the next round of block filter weights , will the new signal template As the old signal template for the next round Return to step 4.

[0076] Because the pulse width of a pulsed laser beam is very short, Most of the data in the waveform is noise, making adjustments to the target waveform meaningless. Therefore, First, the data filtering module needs to be imported. The data filtering module filters data by detecting the amplitude and pulse width characteristics of the signal. Extract valid data The expected data for a block function chain-type adaptive filter.

[0077] Combination Figure 6 The data filtering module works as follows: after the echo signal is sampled in its entirety, it is converted into a discrete sequence inside the FPGA, and then multi-pulse superposition is performed to obtain... At this time peak The corresponding sampling point number is To prevent background noise or spike noise from being misinterpreted as echo signal peaks, the data filtering module needs to detect the echo signal pulse width and amplitude. Define the laser echo signal pulse width. For echo signals greater than The number of sampling points determines the effective data extracted. The following constraints must be met:

[0078] (13),

[0079] In the formula, This represents the average background noise of the laser receiver.

[0080] Combined with the transmitted signal pulse width Then the number of valid data points selected is Therefore, the valid data obtained through the data filtering module is [number]. It should be:

[0081] (14),

[0082] In the formula, express The Middle Each element.

[0083] Combination Figure 3 Block filter weights for linear filtering in the BFLAF algorithm The update algorithm is as follows:

[0084] (15),

[0085] In the formula, valid data As the expectation vector of the block function chain adaptive filter, express The One element, For error signals, express The Each element.

[0086] Then the block filter weights Updated to :

[0087] (16),

[0088] In the formula, This is the filter learning step size.

[0089] The innovation of this invention lies in the use of a data filtering module for filtering the superimposed signals. Preprocessing is performed to filter out non-Gaussian background noise or spike noise in advance, thereby obtaining effective data. As a matched filter, it is used to update the expected data of the signal template, providing a higher quality signal for subsequent data processing. At the same time, the data extension length is determined according to the number of effective data. When the two are the same, the target waveform adaptive adjustment module only needs to run once to complete the update of the target waveform, which greatly improves the real-time processing capability of the algorithm.

[0090] Example 1

[0091] The pulsed laser echo signal enhancement method based on digital signal processing described in this invention was simulated using experimentally measured echo signals and laser emission signals. The pulse width of the laser emission signal was set to... The pulse duration is 20 ns, the pulse repetition frequency is 10 kHz, and the photon number distribution follows a Gaussian distribution. The laser emission signal, after normalization, is as follows: Figure 7 As shown. A 1Gsps ADC is used for sampling. Ideally, the signal template should be the original emission waveform of the laser; therefore, the laser emission waveform is used as the initial old signal template. The pulse width of the laser emission signal is then considered. Data extension block length Number of filter taps The number of valid data points in the data filtering module is also set to 21. This is because the number of valid data points in the data filtering module and the length of the data expansion block are both... If the waveforms are identical, then after each multi-pulse superposition, the target waveform adaptive adjustment module only needs to run once to update the target waveform. Therefore, until the next multi-pulse superposition... Before its arrival, the FPGA had ample runtime to process the nonlinear function extension and weight update parts of the BFLAF algorithm, greatly improving the algorithm's real-time processing capabilities.

[0092] Combination Figure 8 The left figure shows the collected laser echo signal. The target peak is a weak signal, almost submerged by noise, which causes the time identification method to fail. Therefore, the pulse laser echo signal enhancement method based on digital signal processing proposed in this invention must be used to improve the maximum detection range of the underwater laser detection system. After multi-pulse superposition and adaptive matched filtering, the signal-to-noise ratio of the target peak is significantly enhanced. At this time, the time identification algorithm can effectively identify the timing endpoint.

[0093] Combination Figure 9 The echo signal from close-range detection is input into the pulsed laser echo signal enhancement method based on digital signal processing designed in this invention. At this time, the target peak intensity is much greater than the noise floor of the receiving system. Before and after filtering by the method of this invention, the intensity of the target peak is significantly improved. However, the phase of the target peak shifts under the action of the matched filter. This is because the output signal of the matched filter is essentially the correlation between the input signal and the signal template. Therefore, the target peak will exhibit a certain degree of distortion under the action of the matched filter, thus introducing time discrimination error. Therefore, when the target peak amplitude is sufficiently strong, the signal output from the echo signal enhancement method should not be used for time discrimination.

Claims

1. A method for enhancing pulsed laser echo signals based on digital signal processing, characterized in that, The steps are as follows: Step 1: Obtain the laser echo signal based on full waveform sampling: The laser emitter directs a collimated beam onto an underwater target, while the laser receiver captures all diffusely reflected photons within its field of view, acquiring the laser echo signal based on full-waveform sampling. Proceed to step 2; Step 2: Combine the three consecutive laser echo signals acquired as a group and perform multi-pulse superposition to obtain the superimposed signal. Proceed to step 3; Step 3: Design a block function chain adaptive filter, using the laser emission pulse waveform as the initial old signal template. Proceed to step 4; Step 4: Replace the old signal template Input extended data block to obtain extended matrix The nonlinear extended matrix is ​​obtained by extending it with a nonlinear function. Combined with block filter weights After parallel-to-serial conversion, a new signal template for the matched filter is obtained. Finally, the new signal template The signal obtained after superposition of multiple pulses Perform matched filtering to obtain the enhanced output signal. Proceed to step 5; Step 5: Combine the signals after multi-pulse superposition Import data filtering module, from Extract valid data As expected data for block function chain-type adaptive filters, to update the block filter weights. , will the new signal template As the old signal template for the next round Return to step 4.

2. The pulsed laser echo signal enhancement method based on digital signal processing according to claim 1, characterized in that, Step 2 is detailed as follows: Laser echo signal acquired based on full waveform sampling Make the target echo as Random noise is Then we have: (1)， Subscript Indicates the corresponding segment number; The echo signal after m multi-pulse superposition for: (2)， In equation (2), Indicates the first Segment laser echo signal, Indicates the first Target echo in segment laser echo signal, Indicates the first Random noise in segment laser echo signals.

3. The pulsed laser echo signal enhancement method based on digital signal processing according to claim 2, characterized in that, Step 3 is detailed as follows: The block function chain adaptive filter is implemented by using a nonlinear function extension module based on trigonometric functions, combined with the BFLAF algorithm. The functional expression for the nonlinear extension of trigonometric functions is: ,in Let be the number of filter taps in the extension module, and also the number of functions in the extension module, then the filter's tap number is... The nonlinear extension expression for each tap is: (7)， in, Indicates old signal template The Middle One element, , To expand the index, Indicates the length of the discrete sequence. , Since the number of functions in the extension module is Then the extended order for: (8)。 4. The pulsed laser echo signal enhancement method based on digital signal processing according to claim 3, characterized in that, In step 4, the old signal template is... Input extended data block to obtain extended matrix The nonlinear extended matrix is ​​obtained by extending it with a nonlinear function. The details are as follows: The laser emission pulse waveform is used as the initial old signal template. and import extended data blocks. middle, For one OK, The data matrix of columns, where To extend the length of the data, This represents the number of filter taps. Indicates the number of iterations. Initially 0; Let the row vector of the data block be , ,but (6)， Where T represents transpose. express The Middle One element; Let the old signal template be input After expansion using a nonlinear function, the column expansion matrix is ​​obtained. : (9)， The obtained extended data matrix is ​​then extended using a nonlinear function to obtain a nonlinear extended matrix. for: (10)。 5. The pulsed laser echo signal enhancement method based on digital signal processing according to claim 4, characterized in that, In step 4, the block filter weights are combined. After parallel-to-serial conversion, a new signal template for the matched filter is obtained. Finally, the new signal template The signal obtained after superposition of multiple pulses Perform matched filtering to obtain the enhanced output signal. The details are as follows: New signal template for matched filter for: (11)， In the formula, the block filter weights ; The obtained new signal template of the matched filter and the signal after multi-pulse superposition After matched filtering, the enhanced output signal is obtained. : (12)， In the formula, n represents the length of the discrete sequence, and s represents the discrete integral variable.

6. The pulsed laser echo signal enhancement method based on digital signal processing according to claim 5, characterized in that, In step 5, the signal after multi-pulse superposition is... Import data filtering module, from Extract valid data As expected data for block function chain-type adaptive filters, to update the block filter weights. The updated result For the next round of block filter weights , will the new signal template As the old signal template for the next round The details are as follows: The echo signal is sampled in its entirety and converted into a discrete sequence within the FPGA. This sequence is then superimposed using multiple pulses to obtain the final signal. , making this time peak The corresponding sampling point number is ; Define the pulse width of the laser echo signal For echo signals greater than The number of sampling points determines the effective data extracted. The following constraints must be met: (13)， In the formula, This represents the average background noise of the laser receiver; Combined with the transmitted signal pulse width Then the number of valid data points selected is Therefore, the effective data obtained through the data filtering module is [number missing]. It should be: (14)， In the formula, express The Middle One element, Block filter weights The update algorithm is as follows: (15)， In the formula, valid data As the expectation vector of the block function chain adaptive filter, express The One element, For error signals, express The One element; Then the block filter weights Updated to : (16)， In the formula, This is the filter learning step size.

Images