Method and device for image denoising and micro-object recognition based on wavelet transform

Abstract

The embodiment of the present application discloses a kind of based on wavelet transform's image denoising and micro target identification method and device, it is related to computer digital image processing technical field, can save computing resource.The present application includes: the sample image of acquisition is converted, and the wavelet decomposition of image after conversion is carried out, and wavelet coefficient is obtained;Using last layer wavelet coefficient, the background of sample image is reconstructed;Using the background of sample image obtained by reconstruction and the pixel value of sample image, first threshold value is acquired, then first threshold value is used to remove abnormal point in sample image;Sample image is secondly wavelet decomposed, and the low frequency part of last layer wavelet coefficient obtained by second wavelet decomposition is used again, the background of sample image is reconstructed secondly;On the basis of image after removing abnormal point, subtract the background of sample image reconstructed secondly, and according to the second threshold value set to pixel point is handled;Micro target is extracted from target image by connected domain.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer digital image processing, and in particular to a method and device for image denoising and micro-target recognition based on wavelet transform. BACKGROUND

[0002] Target detection is a relatively important research in the field of computer vision, and is also the basis for other complex vision tasks. Small target detection has long been a difficulty in target detection, which aims to accurately detect small targets with few visual features in images. In real-world scenarios, small targets exist in large quantities, so small target detection has broad application prospects and plays an important role in automatic driving, intelligent medical care, defect detection, aerial image analysis and many other fields. However, under normal circumstances, small targets usually lack sufficient appearance information, so it is difficult to distinguish them from the background or similar targets. Real scenes are complex, and there are often problems such as dramatic changes in lighting, target occlusion, densely connected targets, and target size changes, and these factors have a more intense impact on small target features, further increasing the difficulty of small target detection.

[0003] The rapid development of deep learning technology has injected new blood into small target detection, making it a research hotspot. However, deep learning requires a large number of data samples to train the network. Moreover, in order to locate the position of the target in the image, most existing methods generate a series of anchor boxes in advance at each position in the image. However, this approach leads to an imbalance in the number of positive samples for different sizes of targets during model training. When the anchor box set by humans differs greatly from the true bounding box of a small target, the number of training positive samples for the small target will be much smaller than that for large / middle-scale targets, which will cause the trained model to pay more attention to the detection of large / middle-scale targets and ignore the detection of small targets.

[0004] In order to address the problem of small target detection, some solutions have designed a method to obtain the image background through erosion operation and remove image bad points through dilation operation. After obtaining the binary image, the image is further processed by erosion and dilation algorithms to achieve target recovery. However, this approach has a relatively large amount of calculation, consumes a lot of computing resources, and takes a long time. SUMMARY

[0005] Embodiments of the present application provide a method and device for image denoising and micro-target recognition based on wavelet transform, which can save computing resources.

[0006] To achieve the above object, the embodiments of the present application adopt the following technical solutions:

[0007] In a first aspect, the embodiments of the present application provide a method, which comprises:

[0008] S1, converting a collected sample image, and performing wavelet decomposition on the converted image to obtain wavelet coefficients, wherein the converted image is a normalized gray image;

[0009] S2, reconstructing a background of the sample image by using the last layer of wavelet coefficients;

[0010] S3, obtaining a first threshold value by using the reconstructed background of the sample image and pixel values of the sample image, and removing abnormal points in the sample image by using the first threshold value;

[0011] S4, performing second wavelet decomposition on the sample image obtained after S3, and secondly reconstructing the background of the sample image by using the low-frequency part of the last layer of wavelet coefficients obtained by the second wavelet decomposition;

[0012] S5, on the basis of the image after removing the abnormal points in S3, subtracting the background of the sample image secondly reconstructed in S4, and processing pixel points according to a second threshold value, to obtain a target image, wherein the processing of the pixel points includes setting pixel points smaller than the second threshold value to zero and setting pixel points larger than the second threshold value to one;

[0013] S6, extracting a micro target from the target image by using a connected domain.

[0014] In a second aspect, an apparatus provided by an embodiment of the present application includes:

[0015] a processing module configured to convert a collected sample image, and perform wavelet decomposition on the converted image to obtain wavelet coefficients, wherein the converted image is a normalized gray image;

[0016] a first reconstruction module configured to reconstruct a background of the sample image by using the last layer of wavelet coefficients;

[0017] a first filtering module configured to obtain a first threshold value by using the reconstructed background of the sample image and pixel values of the sample image, and remove abnormal points in the sample image by using the first threshold value;

[0018] a second reconstruction module configured to perform second wavelet decomposition on the sample image obtained after S3, and secondly reconstruct the background of the sample image by using the low-frequency part of the last layer of wavelet coefficients obtained by the second wavelet decomposition;

[0019] a second filtering module, configured to subtract a background of the sample image for a second time based on the image after the abnormal points are removed, and process the pixel points according to a second threshold value, to obtain a target image, wherein the processing of the pixel points comprises: setting the pixel points smaller than the second threshold value to zero, and setting the pixel points larger than the second threshold value to one;

[0020] an extraction module, configured to extract the micro target from the target image through a connected domain.

[0021] The method and device for image denoising and micro target recognition based on wavelet transform provided by the embodiments of the present application select a suitable wavelet basis to perform wavelet decomposition on the fuzzy gray image converted into double precision, to obtain wavelet coefficients of each layer. Then, the wavelet coefficients of the last layer after decomposition, i.e. the low-frequency part corresponding to the image, are reconstructed by wavelet to obtain the approximate background of the sample image. The threshold value is set by using the approximate background to remove possible abnormal points, and image decomposition and reconstruction are performed again to obtain the background. Then, the sample image is subtracted by the background to obtain the image after the background is removed. Finally, a suitable threshold value is selected to identify the target from the processed image by using the connected domain method. In the embodiments, the background is reconstructed by using the wavelet coefficients of the last layer of the wavelet transform, i.e. the low-frequency part of the image, a suitable threshold value is set after the background is subtracted, and the image that may be the target is obtained, and the small area object is removed by using the connected domain method. The good target recognition result is obtained without the need of a large number of sample numbers. Compared with the conventional fuzzy image background removal and target recognition method, the design idea of the embodiments can better realize the nonlinear background removal and better preserve the original information of the image. The wavelet basis selection is relatively flexible, and the multi-band wavelet and wavelet packet can be selected according to the signal characteristics and the denoising requirements. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0023] Figure 1 The method flowchart provided by the embodiments of the present application;

[0024] Figure 2 The fuzzy image provided by the embodiments of the present application;

[0025] Figure 3 The reconstructed image background provided by the embodiments of the present application;

[0026] Figure 4 The image after the background is subtracted provided by the embodiments of the present application;

[0027] Figure 5 The target recognition result drawing obtained by using the connected domain after selecting the appropriate threshold value is shown in the embodiment of the present application.

[0028] Figure 6 、 Figure 7 The specific example provided in the embodiment of the present application is shown in the drawing. DETAILED DESCRIPTION

[0029] In order for those skilled in the art to better understand the technical solutions of the present application, the present application will be further described in detail below in combination with the drawings and specific embodiments. The embodiments of the present application will be described in detail below, and examples of the embodiments are shown in the drawings, in which the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are only used to explain the present application, and cannot be interpreted as a limitation on the present application. Those skilled in the art can understand that, unless specifically stated, the singular forms "a", "an" and "the" used herein can also include the plural forms. It should be further understood that the phrase "comprising" used in the specification of the present application means that the features, integers, steps, operations, elements and / or components are present, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we say an element is "connected" or "coupled" to another element, it can be directly connected or coupled to the other element, or there can be intermediate elements. In addition, "connected" or "coupled" used herein can include wireless connection or coupling. The phrase "and / or" used herein includes any one of the associated listed items and all combinations of the associated listed items. Those skilled in the art can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as that generally understood by those skilled in the art in the field to which the present application belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have meanings consistent with those in the context of the prior art, and unless defined as such, should not be interpreted in an idealized or overly formal sense.

[0030] In the field of image processing, wavelet transform is attracting more and more attention in the field of image denoising due to its good time-frequency characteristics, opening up the way for nonlinear denoising. The advantages of wavelet transform processing are: low entropy. The sparse distribution of wavelet coefficients reduces the entropy of the transformed image. That is, after decomposing the signal (i.e. image), more wavelet basis coefficients tend to 0 (noise), while the main part of the signal is concentrated in certain wavelet bases. Threshold denoising can better preserve the original signal. Multi-resolution characteristic. Since a multi-resolution method is used, the non-stationarity of the signal, such as sudden changes and breakpoints, can be well described, and noise can be removed at different resolutions according to the distribution of the signal and noise. De-correlation. Wavelet transform can de-correlate the signal, and the noise tends to be white after transformation, so the wavelet domain is more suitable for denoising than the time domain. Flexible base function selection. Wavelet transform can flexibly select base functions, and can select multi-band wavelets and wavelet packets according to the characteristics of the signal and the denoising requirements for different occasions.

[0031] Therefore, the embodiment aims at the above problems and the advantages of wavelet transform in image processing, and proposes a detection method for removing the background of an image based on wavelet transform to obtain better target recognition results. It has strong practical value. The general design idea is to convert the fuzzy gray image into a double-precision fuzzy gray image, select a suitable wavelet basis for wavelet decomposition, and obtain wavelet coefficients of each layer. Then, the last layer of wavelet coefficients after decomposition, i.e. the low-frequency part of the image, is reconstructed by wavelet to obtain the approximate background of the sample image, and the threshold is set using the approximate background to remove possible abnormal points. Image decomposition and reconstruction are performed again to obtain the background. Then, the sample image is subtracted from the background to obtain an image after removing the background. Finally, a suitable threshold is selected to identify the target from the processed image using the connected domain method.

[0032] Among them, on the basis of wavelet transform, the last layer of wavelet coefficients of the image, i.e. the low-frequency part of the image, is reconstructed to obtain the background, a suitable threshold is set after subtracting the background, and the image that may be the target is obtained. The connected domain method is used to remove small area objects. A good target recognition result is obtained without the need for a large number of sample numbers. Compared with the traditional fuzzy image background removal and target recognition method, the design idea of the embodiment can better achieve nonlinear background removal and better preserve the original information of the image. Moreover, the wavelet basis selection is flexible, and multi-band wavelets and wavelet packets can be selected according to the characteristics of the signal and the denoising requirements.

[0033] The embodiment of the application provides a method for image denoising and micro target recognition based on wavelet transform, as shown in Figure 1 , which comprises the following steps.

[0034] S1, converting the collected sample image and performing wavelet decomposition on the converted image to obtain wavelet coefficients.

[0035] The converted image is a normalized gray image.

[0036] S2, reconstructing the background of the sample image by using the last layer of wavelet coefficients.

[0037] S3, obtaining a first threshold value by using the reconstructed background of the sample image and the pixel value of the sample image, and removing abnormal points in the sample image by using the first threshold value.

[0038] S4, performing second wavelet decomposition on the sample image obtained after S3, and reconstructing the background of the sample image by using the low-frequency part of the last layer of wavelet coefficients obtained by the second wavelet decomposition.

[0039] The image reconstructed by the low-frequency information of the last layer is regarded as the background, and the wavelet decomposition realizes nonlinear background removal, thus retaining most of the high-frequency information of the image, that is, the detail part of the image.

[0040] S5, subtracting the background of the sample image obtained by the second reconstruction in S4 from the image obtained after removing the abnormal points in S3, and processing the pixel points according to a second threshold value, to obtain a target image.

[0041] The processing of the pixel points includes setting the pixel points smaller than the second threshold value to zero and setting the pixel points larger than the second threshold value to one. Specifically, the first threshold value is calculated in S3 and is used to remove abnormal points. The second threshold value can be directly set and is used for binary image. In the preferred scheme, the second threshold value is set to 0.055.

[0042] S6, extracting a micro target from the target image by using a connected domain.

[0043] In the embodiment, in S1, wavelet decomposition is performed on the converted image by using a db6 wavelet basis. Wavelet coefficients corresponding to the profile part of the last layer are taken as the last layer wavelet coefficients. The first layer is the sample image. From the second layer, the image information is divided into a detail part and a profile part, and the profile part is involved in wavelet decomposition of the next layer. It should be noted that the "detail part and profile part" in the embodiment are also referred to as "high-frequency part and low-frequency part" in some research teams of other institutions. The high-pass and low-pass filters are equivalent to convolution of the original image and the corresponding decomposition quantity. The decomposition quantity of each wavelet basis is different. For example, the db6 wavelet basis selected in the embodiment has a high-pass and low-pass decomposition quantity of a 1*12 sequence. The specific value can be viewed by using the wfilter function. Each convolution divides the image into a detail signal and a profile signal (rough signal), that is, a high-frequency signal and a low-frequency signal. The so-called high frequency and low frequency are also relative.

[0044] In a preferred scheme, the number of wavelet decompositions is 7, and a total of 7 layers are obtained. The wavelet coefficients corresponding to the profile information after the seventh image decomposition are obtained in the last layer. The image reconstructed by using the wavelet coefficients corresponding to the profile information of the seventh layer can be taken as the first background (that is, the background of the reconstructed sample image). The reconstruction process is opposite to the decomposition process. In matlab, the reconstruction is performed by using the encapsulated function wrcoef function.

[0045] In the embodiment, in the process of each wavelet decomposition on the converted image, convolution is performed on the high-pass decomposition quantity and the low-pass decomposition quantity corresponding to the wavelet basis, where S*H i =cD i , S*L i =cA i , S represents the converted image, H i represents the high-pass decomposition quantity, L i represents the low-pass decomposition quantity, * represents convolution operation, cD i represents a detail signal, and cA irepresents the sketch signal, i is a positive integer and i≥1. Specifically, the image is decomposed by using db6 wavelet base. The dbN series wavelet has good regularity, and the smooth error introduced as a sparse base is not easy to be detected, so that the signal reconstruction process is relatively smooth. The larger N is, the better the smoothness is, but the calculation amount is greatly increased, and the real-time performance is deteriorated. In the embodiment, the image is decomposed seven times by using db6 wavelet base. The first layer is a sample image, and each time of layering divides the image information into a detail part and a sketch part (or called a high-frequency part and a low-frequency part), that is, the detail information and the sketch information of the image. For example, taking three-layer decomposition as an example, the sample image S (512*512) is decomposed once by using a wavelet base, that is, the high-pass (H i ) low-pass decomposition quantity (L o ) is convolved:

[0046] S*H1=cD1

[0047] S*L1=cA1

[0048] Wherein, * in the formula represents convolution operation, and a detail signal cD1 and a sketch signal cA1 are obtained. Since it is down-sampling, cD1 and cA1 obtained are only half (256*256) of the original S. Similarly, cA1 is further decomposed to obtain cD2 and cA2, and cA2 is continuously decomposed to obtain cD3 and cA3. At this time, three-layer decomposition is completed, and cA3 can be selected for reconstruction to obtain a first background B1. The multi-layer decomposition is also similar operation.

[0049] In the embodiment, the image reconstructed by using the wavelet coefficient corresponding to the sketch information of the seventh layer is taken as the first background. The reconstruction process is opposite to the decomposition process, and the function wrcoef function is encapsulated in matlab to perform reconstruction. Each layer of wavelet decomposition is a process of passing through a high-pass filter and a low-pass filter. That is, the convolution process of the decomposition quantity corresponding to the wavelet, and the result after convolution is the detail signal and the sketch signal corresponding to each layer, wherein the wavelet coefficient is included. The detail and sketch information of the image is retained in the wavelet coefficient. When reconstruction is performed, the wavelet coefficient is used to reconstruct the image required by using the corresponding algorithm.

[0050] In S3, the absolute value of the pixel value of the sample image is square-rooted, divided by two, and then added to the background of the reconstructed sample image to obtain the first threshold value.

[0051] The first threshold value is used to remove the abnormal points in the sample image. For example:

[0052]

[0053] T i =E i +B i

[0054] wherein S i represents the pixel value of the original image, T i represents the threshold value, B i represents the pixel value of the first background reconstructed in the above step. If the pixel value of the original image E i is greater than T i , the corresponding value is changed to T i , and the pixel value smaller than T i is unchanged, and i is a positive integer and i≥1.

[0055] On the basis of the image in which the abnormal points are removed in S3, the second background of the sample image is subtracted in S4, for example, the image in which the abnormal points are removed is decomposed and reconstructed again to obtain a second background B2, and the sample image is subtracted from the second background to obtain only the detail part of the sample image, thereby obtaining an image S1 i which may be the target. In the image, there are still many non-targets, and the pixel points smaller than the threshold value are set to zero and the rest are set to one. The non-target points are removed again. For example, S1 i =S i -B2 i , S i represents the image in which the abnormal points are removed in S3, and B2 i represents the second background of the sample image obtained in S4. S1 i is binarized, and the threshold value is generally selected as 0.055. In S1 i , the pixel points smaller than the threshold value are set to zero and the rest are set to one, thereby obtaining the target image.

[0056] In this embodiment, S6 includes:

[0057] Small area objects in the target image are deleted by using a connected domain, and these small area objects are generally some pixel points which are not removed completely and affect the judgment of the micro target. These small area objects are generally relatively isolated pixel points, and can be deleted by using the method of connected domain. The whole image is traversed from left to right and from top to bottom, and the adjacent pixel values of one object are marked as one,

[0058] For example, the small area objects include target objects in which the number of adjacent pixel points in an 8-adjacent region of 3*3 is less than 3. For example, as shown in the figure, the two connected pixel points in the upper right corner of the figure are not regarded as small targets and will be removed after this step. Figure 6

[0059] For example, as shown in the figure, the two connected pixel points in the upper right corner of the figure are not regarded as small targets and will be removed after this step.

[0059] For example, as shown in the figure, the two connected pixel points in the upper right corner of the figure are not regarded as small targets and will be removed after this step. Figure 7As shown in the wavelet decomposition 1-6 layers, the image background reconstructed by using the profile information can be seen that with the increase of the decomposition times, the details of the image are less and the image tends to be the real background. After subtracting the background from the original sample image, the detail information can be better preserved. In the existing technical solution, the image background is obtained by performing an erosion operation, while in the embodiment, the image background is obtained by wavelet decomposition and reconstruction. The way to remove the bad points in the image in the comparison file is to perform an inflation operation, while in the embodiment, the bad points are removed by using a threshold, and the calculation method of the embodiment is more simple and occupies relatively low computing resources. In the existing solution, after obtaining the binary image without background, the image is further subjected to erosion and inflation algorithm to realize target recovery, and finally the connected domain algorithm is used to extract the micro target. In the embodiment, the connected domain method is used to extract the micro target after obtaining the binary image, so the calculation process is shorter and the efficiency of extracting the micro target is higher. In summary, the embodiment further improves the processing method of removing the background of the image and preserving the details of the image, and further improves the calculation efficiency.

[0060] The embodiment also provides an image denoising and micro target recognition device based on wavelet transform, which comprises:

[0061] A processing module is configured to convert the collected sample image and perform wavelet decomposition on the converted image to obtain wavelet coefficients, wherein the converted image is a normalized gray image.

[0062] A first reconstruction module is configured to reconstruct the background of the sample image by using the last layer of wavelet coefficients.

[0063] A first filtering module is configured to obtain a first threshold value by using the reconstructed background of the sample image and the pixel value of the sample image, and remove abnormal points in the sample image by using the first threshold value.

[0064] A second reconstruction module is configured to perform second wavelet decomposition on the sample image obtained after S3, and reconstruct the background of the sample image for the second time by using the low-frequency part of the last layer of wavelet coefficients obtained by the second wavelet decomposition.

[0065] A second filtering module is configured to subtract the background of the sample image reconstructed for the second time from the image after removing the abnormal points, and process the pixel points according to a set second threshold value to obtain a target image, wherein the processing of the pixel points comprises: setting the pixel points smaller than the second threshold value to zero and setting the pixel points larger than the second threshold value to one.

[0066] An extraction module is configured to extract a micro target from the target image by using a connected domain.

[0067] The processing module is specifically configured to perform wavelet decomposition on the converted image by using a db6 wavelet base, wherein the first layer is the sample image, and image information is divided into a detail part and a sketch part from the second layer, and the sketch part participates in wavelet decomposition of the next layer. The wavelet coefficient corresponding to the sketch part of the last layer is taken as the last layer wavelet coefficient.

[0068] The extraction module is specifically configured to delete small area objects in the target image by using a connected domain, wherein the small area objects include target objects with less than 3 adjacent pixel points in an 8-adjacent region of 3*3.

[0069] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each of the embodiments focuses on the difference from other embodiments. In particular, for the device embodiments, since they are basically similar to the method embodiments, they are described more simply, and the relevant parts can be referred to the part of the method embodiments. The above is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any skilled person in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for image denoising and micro-object identification based on wavelet transform, characterized in that, The method comprises the following steps: S1, converting the collected sample image, and performing wavelet decomposition on the converted image to obtain wavelet coefficients, wherein the converted image is a normalized gray image; S2, reconstructing the background of the sample image by using the last layer of wavelet coefficients; S3, obtaining a first threshold value by using the reconstructed background of the sample image and the pixel value of the sample image, and removing abnormal points in the sample image by using the first threshold value; S4, performing second wavelet decomposition on the sample image obtained after S3, and secondly reconstructing the background of the sample image by using the low-frequency part of the last layer of wavelet coefficients obtained by the second wavelet decomposition; S5, on the basis of the image after removing the abnormal points in S3, subtracting the background of the sample image obtained in S4, and processing the pixel points according to a second threshold value, to obtain a target image, wherein the processing of the pixel points comprises: setting the pixel points smaller than the second threshold value to zero, and setting the pixel points larger than the second threshold value to one; S6, extracting micro targets from the target image by using a connected domain. The removing of the abnormal points in the sample image by using the first threshold value comprises: T i = E i + B i where S i represents the pixel value of the original sample image collected, T i represents a threshold value, B i represents the pixel value of the first background reconstructed in the above step, if the pixel value of the original sample image E i collected is greater than T i , the corresponding value is changed to T i , and the pixel value smaller than T i is unchanged, and i is a positive integer and i≥1.

2. The method of claim 1, wherein, In S1, the method comprises the following steps: performing wavelet decomposition on the converted image by using a db6 wavelet basis, wherein the first layer is the sample image, and the image information is divided into a detail part and a sketch part from the second layer, and the sketch part participates in the wavelet decomposition of the next layer; and the wavelet coefficients corresponding to the sketch part of the last layer are used as the last layer of wavelet coefficients.

3. The method of claim 2, wherein, The number of wavelet decompositions is 7, and a total of 7 layers are obtained.

4. The method according to claim 2 or 3, characterized in that, In the process of each wavelet decomposition on the converted image, the method comprises the following steps: convolution of the high-pass decomposition quantity and the low-pass decomposition quantity corresponding to the wavelet basis, wherein: S*H i = cD i , S*L i = cA i , S represents the converted image, H i represents the high-pass decomposition quantity, L i represents the low-pass decomposition quantity, * represents the convolution operation, cD i represents the detail signal, cA i represents the approximation signal, and i is a positive integer and i≥1.

5. The method of claim 1, wherein, In S3, the method comprises the following steps: taking the square root of the absolute value of the pixel value of the sample image, dividing by two, and then adding the reconstructed background of the sample image to obtain the first threshold value; and removing the abnormal points in the sample image by using the first threshold value.

6. The method of claim 1, wherein, In S6, the method comprises the following steps: deleting small area objects in the target image by using a connected domain, wherein the small area objects comprise target objects with less than 3 adjacent pixel points in an 8-adjacent region of 3*3.

7. A device for image denoising and micro-object identification based on wavelet transform, characterized in that, The method comprises the following steps: a processing module, configured to convert a collected sample image, and perform wavelet decomposition on the converted image to obtain wavelet coefficients, wherein the converted image is a normalized gray image; a first reconstruction module, configured to reconstruct the background of the sample image by using the last layer of wavelet coefficients; a first filtering module, configured to obtain a first threshold value by using the reconstructed background of the sample image and the pixel value of the sample image, and remove abnormal points in the sample image by using the first threshold value; a second reconstruction module, configured to perform second wavelet decomposition on the sample image obtained after S3, and secondly reconstruct the background of the sample image by using the low-frequency part of the last layer of wavelet coefficients obtained by the second wavelet decomposition; A second filtering module is configured to subtract the background of the sample image reconstructed for the second time from the image after removing the abnormal points, and process the pixel points according to a second threshold value, to obtain a target image, wherein the processing of the pixel points includes: setting the pixel points smaller than the second threshold value to zero, and setting the pixel points larger than the second threshold value to one; An extraction module is configured to extract a micro target from the target image through a connected domain. The removing of the abnormal points in the sample image by using the first threshold value includes: T i = E i + B i where S i represents the pixel value of the original sample image collected, T i represents a threshold value, B i represents the pixel value of the first background reconstructed in the above step, if the pixel value of the original sample image E i collected is greater than T i , the corresponding value is changed to T i , and the pixel value less than T i is unchanged, and i is a positive integer and i≥1.

8. The apparatus of claim 7, wherein, The processing module is specifically configured to perform wavelet decomposition on the converted image by using a db6 wavelet base, wherein the first layer is the sample image, and the image information is divided into a detail part and a profile part from the second layer, and the profile part participates in the wavelet decomposition of the next layer; and the wavelet coefficient corresponding to the profile part of the last layer is taken as the last layer wavelet coefficient.

9. The apparatus of claim 7, wherein, The extraction module is specifically configured to delete a small area object in the target image by using a connected domain, wherein the small area object includes a target object with a number of adjacent pixel points in a 3*3 eight-adjacent region less than 3.

Images