Intelligent recognition system for lung nodules based on thoracoscope images

Abstract

The application relates to a thoracoscope image-based lung nodule intelligent identification system. The system comprises a separation module for separating a thoracoscope image into multiple channel images; an enhancement module for performing image enhancement on a first channel image by using multiple enhancement algorithms and performing multiple times of downsampling respectively to obtain corresponding multi-layer scale images; an iteration module for performing iteration processing on the multi-layer scale images to obtain a first target image; the a-th iteration comprises the following steps: analyzing a target scale image of the a-th layer, a second scale image, a third scale image and multiple enhancement scale images to obtain a correction coefficient of the a+1-th layer, correcting a first scale image of the a+1-th layer according to the correction coefficient to obtain a target scale image of the a+1-th layer; and an input module for inputting a second target image into a trained neural network model to obtain an analysis result. The system dynamically weights the advantages of different enhancement algorithms, optimizes the image enhancement effect and improves the accuracy of nodule identification.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a lung nodule intelligent identification system based on thoracoscope images. BACKGROUND

[0002] In recent years, artificial intelligence technology has rapidly emerged in the field of medical image analysis and has gradually been applied to thoracoscope image analysis. Deep learning models, especially convolutional neural networks (CNN) and their improved variants, have been widely used in lung nodule feature analysis. AI-assisted diagnosis systems can quickly process large amounts of thoracoscope image data, reduce missed and misdiagnosed cases, and bring new breakthroughs in the feature analysis and identification of lung nodules under thoracoscopy. Before using neural networks to identify lung nodules, image enhancement is an indispensable preprocessing step. Due to the influence of various factors on the original thoracoscope image, such as uneven illumination, low contrast, smoke obstruction, tissue reflection, and differences in imaging angle and distance, the features of the lung nodule region are not prominent enough, which will directly affect the feature analysis effect of the neural network on the nodule part. Image enhancement can improve the quality of thoracoscope images, enhance the edge, contrast, and texture features of the nodule region, making the nodule more clear and visible, thereby providing better input data for the training and detection of neural networks and helping to improve the performance of the model. However, traditional single image enhancement methods, although they can improve the visual effect of thoracoscope images to some extent, have obvious shortcomings and are difficult to adapt to the diverse needs of different thoracoscope images and nodule features, resulting in less than ideal enhancement effects and failing to meet the requirements of lung nodule analysis of thoracoscope images. SUMMARY

[0003] In order to solve the technical problem of inaccurate single image enhancement algorithm, the purpose of the present application is to provide a lung nodule intelligent identification system based on thoracoscope images, and the technical solution adopted is as follows:

[0004] A lung nodule intelligent identification system based on thoracoscope images is provided, and the system comprises:

[0005] A separation module is used to separate the thoracoscope image into multiple channel images; the multiple channel images include a first channel image, a second channel image, and a third channel image;

[0006] An enhancement module is used to use multiple enhancement algorithms to perform image enhancement on the first channel image, generate multiple enhanced images, and perform multiple downsampling on each enhanced image and the multiple channel images respectively to obtain their corresponding multi-layer scale images;

[0007] The iterative module is used to iteratively process the multi-scale images corresponding to each enhanced image and multiple channel images to obtain the first target image. The a-th iteration in the iterative processing includes: analyzing the target scale image, second scale image, third scale image and multiple enhanced scale images of layer a to obtain the correction coefficients of layer a+1; correcting the first scale image of layer a+1 according to the correction coefficients of layer a+1 to obtain the target scale image of layer a+1; where a is a positive integer, when a=1, the target scale image of layer a is the first scale image of layer a, and the first target image is the target scale image of the last layer; the correction coefficients of layer a+1 indicate the influence of each enhancement algorithm on each pixel in the target scale image of layer a+1.

[0008] The input module is used to input the second target image into the trained neural network model to obtain the analysis results; the second target image is obtained based on the first target image.

[0009] Optionally, the first channel image is a brightness channel image, the second channel image is a hue channel image, and the third channel image is a saturation channel image.

[0010] Optionally, the step of analyzing the target-scale image, the second-scale image, the third-scale image, and multiple enhanced-scale images of layer a to obtain the correction coefficients of layer a+1 includes:

[0011] The correction coefficients for layer a+1 are obtained by analyzing multiple difference images between the target scale image and multiple enhancement scale images in layer a, the pixel value difference of the first target point and its surrounding area in the second scale image of layer a, and the pixel value difference of the second target point and its surrounding area in the third scale image of layer a. The position of the first target point in the second scale image corresponds one-to-one with the position of the second target point in the third scale image.

[0012] In the same multi-scale image, the scale of the a-th scale image is larger than the scale of the (a+1)-th scale image; the target scale image and the first scale image are scale images in the multi-scale image corresponding to the first channel image, the second scale image is the scale image in the multi-scale image corresponding to the second channel image, the third scale image is the scale image in the multi-scale image corresponding to the third channel image, and the multiple enhanced scale images are scale images in the multi-scale image corresponding to each enhanced image.

[0013] Optionally, the step of analyzing multiple difference images between the target-scale image and multiple enhanced-scale images of layer a, the pixel value difference of the first target point and its surrounding area in the second-scale image of layer a, and the pixel value difference of the second target point and its surrounding area in the third-scale image of layer a, to obtain the correction coefficients of layer a+1, includes:

[0014] Based on multiple difference images between the target scale image and multiple enhancement scale images at layer a, multiple enhancement coefficients of multiple enhancement algorithms at layer a are determined; the enhancement coefficients characterize the enhancement capabilities of different enhancement algorithms at different scales.

[0015] Based on the pixel value difference between the first target point and its surrounding area in the second-scale image of layer a, and the pixel value difference between the second target point and its surrounding area in the third-scale image of layer a, the conduction suppression coefficient of the target point in the target-scale image of layer a is determined; wherein, the position of the target point in the first-scale image corresponds one-to-one with the position of the first target point in the second-scale image.

[0016] We analyze multiple enhancement coefficients of various enhancement algorithms in layer a and the transmission suppression coefficients of target points in the target-scale image of layer a to obtain the correction coefficients of layer a+1.

[0017] Optionally, the step of determining the transmission inhibition coefficient of the target point in the target-scale image of layer a based on the pixel value difference between the first target point and its surrounding area in the second-scale image of layer a, and the pixel value difference between the second target point and its surrounding area in the third-scale image of layer a, includes:

[0018] The first conduction suppression coefficient of the first target point is determined based on the absolute difference between the pixel value of the first target point and the mean pixel value of the first target region in the second-scale image of the a-th layer; the first target region is determined based on the first target point.

[0019] The second conduction suppression coefficient of the second target point is determined based on the absolute difference between the pixel value of the second target point and the mean pixel value of the second target region in the third-scale image of the a-th layer; the second target region is determined based on the second target point.

[0020] The conduction suppression coefficient of the target point in the target-scale image of layer a is determined based on the average of the first conduction suppression coefficient and the second conduction suppression coefficient.

[0021] Optionally, the step of analyzing multiple enhancement coefficients of multiple enhancement algorithms at layer a and the transmission suppression coefficients of target points in the target-scale image at layer a to obtain the correction coefficients at layer a+1 includes:

[0022] Based on multiple enhancement coefficients of multiple enhancement algorithms in layer a, multiple dominant coefficients of multiple enhancement algorithms in layer a, the transmission suppression coefficient of the target point in the target-scale image of layer a, and the local weights of the target point in the target-scale image of layer a, the transmission weights of the target point in the target-scale image of layer a are determined; wherein, the local weights characterize the influence of multiple enhancement algorithms on the target point in the target-scale image of layer a; the dominant coefficient of each enhancement algorithm in layer a is obtained based on the enhancement coefficient of the corresponding enhancement algorithm in layer a.

[0023] The transmission weights of target points in the target-scale image of layer a and the fluctuation weights of the third target point in the first-scale image of layer a+1 are analyzed to obtain the correction coefficients of layer a+1. The fluctuation weights are obtained based on the pixel difference between the target-scale image of layer a and the first-scale image of layer a+1, and the position of the third target point in the first-scale image corresponds one-to-one with the position of the first target point in the second-scale image.

[0024] Optionally, the step of determining the local weights of target points in the target-scale image of multiple enhancement algorithms at layer a includes:

[0025] When a=1, the local weights of the target points in the target scale image of layer a are determined based on the multiple difference images between the target scale image of layer a and the multiple enhanced scale images of layer a, and the average pixel value of the target scale image of layer a.

[0026] When a≥2, the local weights of the target points in the target-scale image of the a-th layer are determined based on the correction coefficients of the (a-1)th layer.

[0027] Optionally, the step of analyzing the propagation weights of target points in the target-scale image of layer a and the fluctuation weights of the third target point in the first-scale image of layer a+1 to obtain the correction coefficients of layer a+1 includes:

[0028] The correction coefficients for layer a+1 are determined by summing the propagation weights of the target points in the target-scale image of layer a and the fluctuation weights of the third target points in the first-scale image of layer a+1.

[0029] Optionally, the step of correcting the first-scale image of the (a+1)th layer according to the correction coefficients of the (a+1)th layer to obtain the target-scale image of the (a+1)th layer includes:

[0030] The target scale image of layer a+1 is determined by multiplying the pixel value of each pixel in the first scale image of layer a+1 with the correction coefficient of layer a+1.

[0031] Optionally, the plurality of enhancement algorithms include at least two of the following: histogram equalization, anisotropic filtering, adaptive histogram truncation, and gamma correction.

[0032] Based on common knowledge in the field, the above-mentioned preferred conditions can be combined arbitrarily to obtain various preferred embodiments of the present invention.

[0033] This application offers the following advantages: It employs a multi-strategy image enhancement technique, and the dynamic weighted fusion method avoids regional characteristic conflicts caused by fixed-weight fusion, better combining the advantages of different enhancement algorithms, further optimizing the image enhancement effect, and improving the accuracy of nodule identification. It effectively improves the quality of thoracic images, enhancing the edge, contrast, and texture features of the nodule region, making the nodules clearer and more visible. By using multiple image enhancement algorithms and flexibly selecting and combining them according to the specific features and needs of the image, it can adapt to the diverse needs of different thoracic images and nodule features, enhancing the system's processing capability for different types of images. The multi-scale analysis and weight transmission mechanism fully consider the structural information of the image at different scales and the performance differences of the enhancement algorithms. By dynamically adjusting the weights, the image enhancement processing becomes more in line with actual needs, improving the system's adaptability to complex image scenes. It provides higher-quality input data for neural network training and detection, thereby significantly improving the model's recognition performance of lung nodules and reducing the possibility of misdiagnosis and missed diagnosis. It can quickly process a large amount of thoracoscopic image data in a short time and automatically identify lung nodules, which greatly improves the diagnostic efficiency of doctors, reduces the workload and time cost of manual screening, and the improved nodule identification accuracy helps doctors to detect lung nodules earlier and more accurately, providing a more reliable basis for patients' treatment decisions, thereby improving the treatment effect and survival rate of patients. It is of great significance for the screening and diagnosis of early lung cancer. Attached Figure Description

[0034] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a schematic diagram of the structure of a lung nodule intelligent recognition system based on thoracoscopic images, provided in one embodiment of this application.

[0036] Figure 2 A flowchart illustrating a method for intelligent identification of lung nodules based on thoracoscopic images, provided as an embodiment of this application;

[0037] Figure 3 This is a schematic diagram of the structure of an electronic device provided in one embodiment of this application. Detailed Implementation

[0038] To further illustrate the technical means and effects adopted by this application to achieve the intended purpose of the invention, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a lung nodule intelligent recognition system based on thoracoscopic images proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0039] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0040] The following description, in conjunction with the accompanying drawings, details a specific scheme for an intelligent lung nodule recognition system based on thoracoscopic images provided in this application. For example... Figure 1 As shown, the intelligent lung nodule recognition system based on thoracoscopic images includes:

[0041] Separation module 11 is used to separate thoracoscopic images into multiple channel images.

[0042] The multiple channel images include a first channel image, a second channel image, and a third channel image.

[0043] Multiple thoracoscopy images are acquired via thoracoscopy. For a single thoracoscopy image, the acquired image can be separated into multiple channel images. These multiple channel images may include a first channel image, a second channel image, and a third channel image. In one embodiment, the thoracoscopy image is converted to the HSV (Hue, Saturation, Value) color space and separated into multiple channel images. The first channel image can be a value channel (V channel) image, the second channel image can be a hue channel (H channel) image, and the third channel image can be a saturation channel (S channel) image. Separating color and brightness makes it more suitable for processing images under different lighting conditions, improving the image's robustness to lighting conditions.

[0044] The enhancement module 12 is used to enhance the first channel image using multiple enhancement algorithms to generate multiple enhanced images, and to downsample each enhanced image and the multiple channel images multiple times to obtain their respective multi-scale images.

[0045] The core challenge in feature analysis and identification of nodule regions lies in distinguishing nodules with low contrast, varied morphologies, and textures similar to surrounding tissues (such as blood vessels, pleura, and lung parenchyma) from a complex background. Therefore, a targeted image enhancement algorithm is selected for image enhancement processing. In one embodiment, the multiple enhancement algorithms include at least two of the following: histogram equalization, anisotropic filtering, adaptive histogram truncation, and gamma correction. Multiple enhancement algorithms are used to perform image enhancement operations on the first channel image, resulting in multiple enhanced images. Histogram equalization can improve local contrast, anisotropic filtering can improve structural clarity, and adaptive histogram truncation and gamma correction can unify brightness and prevent overexposure. Other image enhancement algorithms can be selected according to the actual situation, such as brightness-guided fusion, adaptive gamma correction, hue and saturation perturbation, and CLAHE+ homomorphic filtering.

[0046] Multiple enhancement algorithms are used to enhance the first channel image, resulting in multiple enhanced images. ,in For example, the enhanced image corresponding to the i-th enhancement algorithm. For histogram equalization, This applies to anisotropic filtering, and so on.

[0047] Each enhancement algorithm and multiple channel images are downsampled multiple times to obtain their respective multi-scale images. The Laplacian pyramid can then be used to analyze the multiple enhanced images. Three-level pyramids were constructed to obtain images at three different scales corresponding to multiple augmented images. ,in For the i-th enhancement algorithm, the multi-scale image at the i-th scale is... Layer, where i is the i-th enhancement algorithm. For the first pyramid Layers, within the same multi-scale image, decrease in scale from top to bottom. It can be 1, 2, or 3. The larger the value, the larger the image size and the smaller the scale. When j=3, the image is the original size without downsampling; when j=1, the image is at its largest scale, i.e., the top layer of the pyramid. A three-level pyramid is constructed for each of the multiple channel images to obtain the images at three different scales corresponding to the first channel image. The second channel image corresponds to three layers of images at different scales. The third channel image corresponds to three layers of images at different scales. . The first channel image is the first multi-scale image corresponding to the first channel image. layer, The second channel image is the first multi-scale image corresponding to the second channel image. layer, The third channel image is the first of the multi-scale images corresponding to the third channel image. layer.

[0048] Since different enhancement algorithms focus on image enhancement in different directions, their advantages can be combined to obtain better image enhancement results. However, directly performing a global weighted summation of different enhancement results carries the risk of "multi-modal mixing chaos," especially in scenarios like medical images where structural information is extremely sensitive. Fixed-weight multi-modal fusion can easily lead to conflicts between different enhancement effects, known as regional characteristic conflicts, resulting in poor enhancement outcomes.

[0049] Because different enhancement algorithms focus on different enhancement aspects, their performance varies across different image scales. For example, gamma correction primarily enhances the overall brightness of the image, thus performing better at large scales; anisotropic filtering primarily enhances the distinction between different structures in the image, thus performing better at small scales. To avoid regional characteristic conflicts caused by fixed-weight fusion, different weight combinations can be applied to different regions of the image, thereby minimizing regional characteristic conflicts.

[0050] The iteration module 13 is used to iteratively process the multi-scale images corresponding to each enhanced image and multiple channel images to obtain the first target image. The a-th iteration in the iterative processing includes: analyzing the target scale image, the second scale image, the third scale image and multiple enhanced scale images of the a-th layer to obtain the correction coefficients of the (a+1)-th layer; and correcting the first scale image of the (a+1)-th layer according to the correction coefficients of the (a+1)-th layer to obtain the target scale image of the (a+1)-th layer.

[0051] Where a is a positive integer, when a=1, the target scale image of layer a is the first scale image of layer a, and the first target image is the target scale image of the last layer; the correction coefficient of layer a+1 indicates the influence of each enhancement algorithm on each pixel in the target scale image of layer a+1.

[0052] In the same multi-level scale image, the scale of the a-th level scale image is larger than the scale of the (a+1)-th level scale image. Understandably, the a-th level is closer to the upper level of the pyramid than the (a+1)-th level.

[0053] The target scale image and the first scale image are scale images in the multi-level scale images corresponding to the first channel image. The second scale image is the scale image in the multi-level scale images corresponding to the second channel image. The third scale image is the scale image in the multi-level scale images corresponding to the third channel image. Multiple enhancement scale images are scale images in the multi-level scale images corresponding to each enhancement image. When it is necessary to calculate the correction coefficients for layer j, the target scale image, the second scale image, the third scale image, and multiple enhancement scale images of layer j-1 can be analyzed to obtain the correction coefficients for layer j. Based on the correction coefficients for layer j, the first scale image of layer j-1 is corrected to obtain the target scale image of layer j. This can be done by analyzing the target scale image of layer j-1 in the multi-level scale images corresponding to the V channel image, the second scale image of layer j-1 in the multi-level scale images corresponding to the H channel image, the third scale image of layer j-1 in the multi-level scale images corresponding to the S channel image, and the enhancement scale image of layer j-1 in the multi-level scale images corresponding to each enhancement image, to obtain the correction coefficients for layer j. Based on the correction coefficients for layer j, the first scale image of layer j in the multi-level scale images corresponding to the V channel image is then corrected.

[0054] Each enhanced image and the corresponding multi-scale images of multiple channels obtained above are iteratively processed to obtain the first target image. Based on the top target scale image, the second scale image, the third scale image, and multiple enhanced scale images, correction coefficients of the adjacent layers are obtained to correct the first scale images of the adjacent layers, thus obtaining the target scale images of the adjacent layers. This process is carried out layer by layer until the target scale image of the bottom layer is obtained. Combining the image information of each upper layer and the image information of the adjacent layers, the target scale images of the adjacent layers are determined. This avoids the regional characteristic conflict problem caused by fixed weight fusion, better combines the advantages of different enhancement algorithms, further optimizes the image enhancement effect, and further improves the accuracy of nodule recognition.

[0055] In one embodiment, the step of analyzing the target-scale image, the second-scale image, the third-scale image, and multiple enhanced-scale images of layer a to obtain the correction coefficients of layer a+1 includes:

[0056] The correction coefficients for layer a+1 are obtained by analyzing multiple difference images between the target scale image and multiple enhancement scale images of layer a, the pixel value difference of the first target point and its surrounding area in the second scale image of layer a, and the pixel value difference of the second target point and its surrounding area in the third scale image of layer a. The position of the first target point in the second scale image corresponds one-to-one with the position of the second target point in the third scale image.

[0057] In thoracoscopic images, pulmonary nodules often exhibit different hue and saturation compared to other areas. That is, within the H-channel and S-channel images, pulmonary nodules show differences from adjacent areas, and this difference is often related to the type of nodule. Since the hue and saturation changes caused by nodules compared to other details in the image vary across different scales, multi-scale images are obtained from the H-channel and S-channel images using an image pyramid. Then, the abnormal behavior of each pixel is captured on the large-scale image. The larger the abnormal behavior, the more likely it is to be a nodule region. Image enhancement focuses more on detail information, and the stronger the suppression effect of high-level weighted transmission. When calculating the correction coefficient of the j-th layer, the first target point in the second-scale image of the (j-1)-th layer can be obtained, and mapped to the same position in the second-scale image of the (j-1)-th layer to obtain the second target point. That is, the first target point and the second target point correspond one-to-one. Then, the pixel value differences between the first target point and its surrounding area in the second-scale image of the (j-1)-th layer, and the pixel value differences between the second target point and its surrounding area in the third-scale image of the (j-1)-th layer are analyzed. The correction coefficients for layer j are obtained by considering multiple difference images between the target scale image and multiple enhancement scale images in layer j-1, the pixel value difference of the first target point and its surrounding area in the second scale image of layer j-1, and the pixel value difference of the second target point and its surrounding area in the third scale image of layer j-1.

[0058] In one embodiment, the step of analyzing multiple difference images between the target-scale image and multiple enhanced-scale images of layer a, the pixel value difference of the first target point and its surrounding area in the second-scale image of layer a, and the pixel value difference of the second target point and its surrounding area in the third-scale image of layer a, to obtain the correction coefficients of layer a+1, includes:

[0059] Based on multiple difference images between the target scale image and multiple enhancement scale images at layer a, multiple enhancement coefficients of multiple enhancement algorithms at layer a are determined; the enhancement coefficients characterize the enhancement capabilities of different enhancement algorithms at different scales.

[0060] Based on the pixel value difference between the first target point and its surrounding area in the second-scale image of layer a, and the pixel value difference between the second target point and its surrounding area in the third-scale image of layer a, the conduction suppression coefficient of the target point in the target-scale image of layer a is determined; wherein, the position of the target point in the first-scale image corresponds one-to-one with the position of the first target point in the second-scale image.

[0061] We analyze multiple enhancement coefficients of various enhancement algorithms in layer a and the transmission suppression coefficients of target points in the target-scale image of layer a to obtain the correction coefficients of layer a+1.

[0062] After image enhancement, comparing the enhanced image with the original image reveals differences. The greater the difference, the more significant the effect of the image enhancement method in that region, resulting in a larger enhancement coefficient for the enhanced image in regions of greater difference. When calculating multiple enhancement coefficients of multiple enhancement algorithms at layer j-1, multiple enhanced images can be obtained. Image at target scale respectively By subtracting the values, multiple difference images are obtained. , Characterizing the enhanced image of the i-th algorithm at layer j-1 and the target-scale image For the i-th enhancement algorithm, the mean of the difference images at different scales is obtained. Then, the mean of the difference images at different scales of the i-th enhancement algorithm is calculated. Normalization yields the enhancement coefficients of the i-th enhancement algorithm at layer j-1. , where i is the i-th enhancement algorithm and j-1 is the j-1-th layer of the pyramid.

[0063] Based on the pixel value differences between the first target point and its surrounding area in the second-scale image of the (j-1)th layer of the H-channel image, and the pixel value differences between the second target point and its surrounding area in the third-scale image of the (j-1)th layer of the S-channel image, the conduction suppression coefficient of the target point in the target-scale image of the (j-1)th layer of the V-channel image is determined. The larger the difference, the more likely it is to be a nodule region. Image enhancement focuses more on detail information, hence the larger the conduction suppression coefficient. Multiple enhancement coefficients of various enhancement algorithms in the (j-1)th layer, and the conduction suppression coefficient of the target point in the target-scale image of the (j-1)th layer, are analyzed to obtain the correction coefficients for the j-th layer.

[0064] In one embodiment, the step of determining the conduction suppression coefficient of a target point in the target-scale image of layer a based on the pixel value difference between the first target point and its surrounding area in the second-scale image of layer a, and the pixel value difference between the second target point and its surrounding area in the third-scale image of layer a, includes:

[0065] The first conduction suppression coefficient of the first target point is determined based on the absolute difference between the pixel value of the first target point and the mean pixel value of the first target region in the second-scale image of layer a; the first target region is determined based on the first target point.

[0066] The second conduction suppression coefficient of the second target point is determined based on the absolute difference between the pixel value of the second target point and the mean pixel value of the second target region in the third-scale image of layer a; the second target region is determined based on the second target point.

[0067] The conduction inhibition coefficient of the target point in the target-scale image of layer a is determined based on the average of the first and second conduction inhibition coefficients.

[0068] When it is necessary to calculate the conduction suppression coefficient of the (j-1)th layer, one can first... In the process, the target point k is determined. The target point can be... Any pixel in the array, then map the target point to... and At the same position, the first target point k' and the second target point are obtained. The V-channel image corresponds to the multi-scale image of the first... Layer, specifically the (j-1)th layer, refers to the target-scale image. The first target area is defined with the first target point as the center; for example, a 7×7 area can be set. The first multi-scale image corresponding to the H-channel image Layer, in The second target area is defined with the second target point as the center; for example, a 7×7 area can be set. The first multi-scale image corresponding to the S-channel image -1 floor. Recalculate. The pixel value of the first target point k' and the average pixel value within the first target area The image mean within the first target region is used to represent the pixel values ​​of physiological tissues around the first target point k' in the laparoscopic image: when the pixel value of the first target point k'... The average pixel value within the first target area The greater the difference, the more likely it is to be a nodule region. In this case, more attention should be paid to its local details, and the greater the inhibition coefficient of upper-level weight propagation, the greater the propagation inhibition coefficient of the first target point k'. The calculation formula is as follows: ,in, This represents the conduction inhibition coefficient at the first target point k'; This represents the average pixel value within the first target area; This represents the pixel value of the first target point k'. The conduction inhibition coefficient of the second target point is calculated using the same method as for calculating the conduction inhibition coefficient of the first target point k'. Then, the average of the conduction inhibition coefficients of the first and second target points is calculated to obtain the conduction inhibition coefficient of target point k. .

[0069] In one embodiment, the step of analyzing multiple enhancement coefficients of multiple enhancement algorithms at layer a and the transmission suppression coefficients of target points in the target-scale image at layer a to obtain the correction coefficients at layer (a+1) includes:

[0070] Based on multiple enhancement coefficients of multiple enhancement algorithms in layer a, multiple dominant coefficients of multiple enhancement algorithms in layer a, the transmission suppression coefficient of the target point in the target-scale image of layer a, and the local weights of the target point in the target-scale image of layer a, the transmission weights of the target point in the target-scale image of layer a are determined; wherein, the local weights characterize the influence of multiple enhancement algorithms on the target point in the target-scale image of layer a; the dominant coefficient of each enhancement algorithm in layer a is obtained based on the enhancement coefficient of the corresponding enhancement algorithm in layer a.

[0071] The transmission weights of target points in the target-scale image of layer a and the fluctuation weights of the third target point in the first-scale image of layer a+1 are analyzed to obtain the correction coefficients of layer a+1. The fluctuation weights are obtained based on the pixel difference between the target-scale image of layer a and the first-scale image of layer a+1, and the position of the third target point in the first-scale image corresponds one-to-one with the position of the first target point in the second-scale image.

[0072] When it is necessary to calculate the propagation weight of the target point in the target-scale image of layer j-1, the enhancement coefficient of the i-th enhancement algorithm in layer j-1 can be used. Calculate the dominant coefficients of multiple enhancement algorithms at layer j-1. Since larger-scale images better reflect the overall structural information, to avoid excessive enhancement of information in small-scale images that could disrupt the continuity of tissues and organs, a multi-level weight transmission constraint mechanism can be designed, combining the aforementioned three-layer image pyramid, to obtain appropriate weights. The dominance of different enhancement algorithms varies at different image scales. For example, in large-scale images, more emphasis is placed on the overall brightness of different regions, making the gamma transform more sensitive than other enhancement methods, thus increasing its dominance. Therefore, based on the enhancement coefficients of different algorithms at different layers obtained above, the dominant coefficients of different enhancement methods at each scale are calculated in order from large to small scale. The calculation formula is: ,in, This represents the dominant coefficient of the enhancement algorithm i in layer (j-1). This represents the enhancement coefficient of enhancement algorithm i at layer (j-1). This represents the sum of enhancement coefficients of all enhancement algorithms in layer j-1. Then, based on the multiple enhancement coefficients of multiple enhancement algorithms in layer j-1, the multiple dominant coefficients of multiple enhancement algorithms in layer j-1, the transmission suppression coefficient of the target point in the target-scale image of layer j-1, and the local weights of the target point in the target-scale image of layer j-1, the transmission weights of the target point in the target-scale image of layer j-1 are determined. The calculation method is as follows: Then, analyze the transmission weight of the target point in the target scale image of layer j-1 and the fluctuation weight of the third target point in the first scale image of layer j to obtain the correction coefficient of layer j.

[0073] in: This represents the propagation weight of the i-th enhancement algorithm at the target point k in layer j-1; This represents the local weight of the target point k in the (j-1)th layer of the i-th enhancement algorithm; This represents the dominance coefficient of the i-th enhancement algorithm in layer j-1; This represents the enhancement coefficient of the i-th enhancement algorithm in layer j-1; The larger the transmission inhibition coefficient, the more abnormal the performance of target point k in the H and S channels, and the more attention should be paid to its detailed performance, and the weaker the transmission of higher-level weights.

[0074] In one embodiment, the step of analyzing the propagation weights of target points in the target-scale image of layer a and the fluctuation weights of the third target point in the first-scale image of layer a+1 to obtain the correction coefficients of layer a+1 includes:

[0075] The correction coefficients for layer a+1 are determined by analyzing the sum of the propagation weights of the target points in the target-scale image of layer a and the fluctuation weights of the third target points in the first-scale image of layer a+1.

[0076] Since the mapping relationship between different scales in multi-scale images is one-to-many, meaning one pixel in the (j-1)th scale image corresponds to four pixels in the j-th scale image, obtaining the local weights of each pixel in the j-th scale image requires combining the image information difference between the j-th and (j-1)-th scales. For a target point k in the target scale image of the (j-1)-th scale, obtain the four matching third target points in the first scale image of the j-th scale. Calculate the pixel difference between the third target point in layer j and the target point k corresponding to layer j-1. Calculate pixel difference With the pixel value of the (j-1)th layer The absolute ratio is used as the volatility weight, and the formula for calculating the volatility weight is as follows: ;in, The third target point of the j-th layer Fluctuation weights, The third target point of the j-th layer The pixel difference corresponding to the target point k in the (j-1)th layer, Let k be the pixel value of the target point k in the (j-1)th layer.

[0077] Based on the transmission weight of target point k in the target scale image of layer j-1 and the third target point in the first scale image of layer j... The sum of the fluctuation weights determines the correction coefficients for the j-th layer, which is also the local weights of each pixel in the first-scale image of the j-th layer. The correction coefficients for the j-th layer are thus the local weights. The calculation formula is as follows: ,in, The third target point in the first-scale image of layer j. Local weights, Let be the propagation weight of target point k in the target-scale image of layer j-1. The third target point in the first-size image of layer j. Fluctuation weights.

[0078] In one embodiment, the step of determining the local weights of target points in the target-scale image of multiple enhancement algorithms at layer a includes:

[0079] When a=1, the local weights of the target points in the target scale image of layer a are determined based on the multiple difference images between the target scale image of layer a and the multiple enhanced scale images of layer a, and the average pixel value of the target scale image of layer a.

[0080] When a≥2, the local weights of the target points in the target-scale image of the a-th layer are determined based on the correction coefficients of the (a-1)th layer.

[0081] When j=1, that is, when the target scale image of layer j is the first scale image of layer j, the difference image between the i-th enhancement algorithm and the target scale image of layer 1 can be calculated. Then calculate the average pixel value of the target-scale image in layer 1. Then based on the average pixel value For difference images The local fluctuations in the image are quantized to obtain the local weights of the target points in the target-scale image of the first layer for multiple enhancement algorithms, and the local weights of target point k. The calculation formula is: ,in, This represents the pixel difference between target point k in the target-scale image of layer 1 and the corresponding pixel in the enhanced image of the i-th enhancement algorithm in layer 1. The average pixel value of the target scale image at layer 1. This represents the enhancement coefficient of the i-th enhancement algorithm in layer 1. This represents the local weight of target point k in the target scale image of the i-th enhancement algorithm at layer 1.

[0082] The local weights of the target points in the target-scale image at layer j of multiple enhancement algorithms are also the correction coefficients of layer j. When j ≥ 2, the correction coefficients of layer j can be used to calculate the third target point in the first-scale image at layer j, which is also based on the above calculation. The local weights are calculated by using the local weights method to calculate the local weights of target point k in the target-scale image of the j-th layer by multiple enhancement algorithms.

[0083] In one embodiment, the step of correcting the first-scale image of layer a+1 according to the correction coefficients of layer a+1 to obtain the target-scale image of layer a+1 includes:

[0084] The target scale image of layer a+1 is determined by multiplying the pixel value of each pixel in the first scale image of layer a+1 with the correction coefficient of layer a+1.

[0085] The local weights of all target points k in the first-scale image of layer j are normalized to obtain the final local weights. The target scale image of layer j is determined by multiplying the pixel values ​​of all target points k in the first-scale image of layer j with the correction coefficient of layer j. The calculation formula is as follows: ,in, Let k be the pixel value of the target point k in the target-scale image. This represents the final local weights after normalization by the i-th enhancement algorithm. This represents the pixel value corresponding to the i-th enhancement algorithm on the target scale image, where n is the total number of enhancement algorithms.

[0086] The correction coefficients are calculated iteratively layer by layer according to the above steps, and corrections are performed to obtain the first target image after the weighted processing of the last layer.

[0087] Input module 14 is used to input the second target image into the trained neural network model to obtain analysis results.

[0088] The second target image is obtained from the first target image. The weighted first target image can be combined with the second channel image and the third channel image to reconstruct the second target image.

[0089] The neural network model is pre-trained. The input to the neural network model is a thoracic image labeled with the boundaries and feature information of the nodules. The output analysis results include the edge contour of the nodules, the region they are located in, and feature parameters, which are used to identify the region where the nodules are located and the feature information of the nodules. Providing the feature parameters of the nodules and the region where the nodules are located makes it easier for doctors to identify and analyze them, and helps doctors to make accurate judgments about the condition of the nodules.

[0090] This application employs a multi-strategy image enhancement technique and a dynamic weighted fusion method to avoid regional characteristic conflicts caused by fixed-weight fusion. It better combines the advantages of different enhancement algorithms, further optimizing the image enhancement effect and improving the accuracy of nodule identification. This effectively improves the quality of thoracic images, enhancing the edge, contrast, and texture features of the nodule region, making the nodules clearer and more visible. By using multiple image enhancement algorithms and flexibly selecting and combining them according to the specific features and needs of the image, it can adapt to the diverse needs of different thoracic images and nodule features, enhancing the system's processing capability for different types of images. Multi-scale analysis and weight transmission mechanisms fully consider the structural information of the image at different scales and the performance differences of the enhancement algorithms. Through dynamic weight adjustment, the image enhancement processing is more in line with actual needs, improving the system's adaptability to complex image scenes. This provides higher-quality input data for neural network training and detection, significantly improving the model's recognition performance of lung nodules and reducing the possibility of misdiagnosis and missed diagnosis. It can quickly process a large amount of thoracoscopic image data in a short time and automatically identify lung nodules, which greatly improves the diagnostic efficiency of doctors, reduces the workload and time cost of manual screening, and the improved nodule identification accuracy helps doctors to detect lung nodules earlier and more accurately, providing a more reliable basis for patients' treatment decisions, thereby improving the treatment effect and survival rate of patients. It is of great significance for the screening and diagnosis of early lung cancer.

[0091] This application also provides a method for intelligent identification of pulmonary nodules based on thoracoscopic images, such as... Figure 2 As shown, the method includes:

[0092] S21. Separate the thoracoscopic image into multiple channel images; the multiple channel images include a first channel image, a second channel image, and a third channel image;

[0093] S22. Multiple enhancement algorithms are used to enhance the first channel image, generating multiple enhanced images. Each enhanced image and multiple channel images are downsampled multiple times to obtain their respective multi-scale images.

[0094] S23. Iteratively process the multi-scale images corresponding to each enhanced image and multiple channel images to obtain the first target image. The a-th iteration in the iterative process includes: analyzing the target scale image, second scale image, third scale image and multiple enhanced scale images of layer a to obtain the correction coefficients of layer a+1; correcting the first scale image of layer a+1 according to the correction coefficients of layer a+1 to obtain the target scale image of layer a+1; where a is a positive integer, when a=1, the target scale image of layer a is the first scale image of layer a, and the first target image is the target scale image of the last layer; the correction coefficients of layer a+1 indicate the influence of each enhancement algorithm on each pixel in the target scale image of layer a+1.

[0095] S24. Input the second target image into the trained neural network model to obtain the analysis results; the second target image is obtained based on the first target image.

[0096] For the method embodiments, since they are basically corresponding to the system embodiments, the relevant parts can be referred to in the description of the system embodiments.

[0097] It should be understood that, although Figure 2 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 2 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0098] Figure 3 This is a schematic diagram of the structure of an electronic device according to an example embodiment of this application. The electronic device includes a memory, a processor, and a computer program stored in the memory and used to run on the processor. When the processor executes the computer program, it implements the method described in any of the above embodiments. Figure 3 The electronic device 30 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0099] like Figure 3As shown, the electronic device 30 can be manifested as a general-purpose computing device, such as a server device. The components of the electronic device 30 may include, but are not limited to: at least one processor 31, at least one memory 32, and a bus 33 connecting different system components (including memory 32 and processor 31).

[0100] Bus 33 includes a data bus, an address bus, and a control bus.

[0101] The memory 32 may include volatile memory, such as random access memory (RAM) 321 and / or cache memory 322, and may further include read-only memory (ROM) 323.

[0102] The memory 32 may also include a program tool 325 (or utility) having a set (at least one) program module 324, such program module 324 including but not limited to: an operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0103] The processor 31 executes various functional applications and data processing, such as the methods provided in any of the above embodiments, by running computer programs stored in the memory 32.

[0104] Electronic device 30 can also communicate with one or more external devices 34 (e.g., keyboard, pointing device, etc.). This communication can be performed via input / output (I / O) interface 35. Furthermore, electronic device 30 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public network, such as the Internet) via network adapter 36. As shown, network adapter 36 communicates with other modules of electronic device 30 via bus 33. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with electronic device 30, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.

[0105] It should be noted that although several units / modules or sub-units / modules of the electronic device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0106] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method provided in any of the above embodiments.

[0107] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0108] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0109] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the method described in any of the above embodiments.

[0110] The program code for executing the computer program product of this application can be written in any combination of one or more programming languages. The program code can be executed entirely on the user device, partially on the user device, as a standalone software package, partially on the user device and partially on a remote device, or entirely on a remote device.

[0111] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0112] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application.

[0113] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A thoracoscope image-based lung nodule intelligent identification system, characterized in that, The system comprises: a separation module for separating the thoracoscope image into a plurality of channel images; the plurality of channel images comprises a first channel image, a second channel image and a third channel image; an enhancement module for performing image enhancement on the first channel image by using a plurality of enhancement algorithms to generate a plurality of enhanced images, and performing multiple times of down-sampling on each enhanced image and the plurality of channel images respectively to obtain a plurality of multi-layer scale images corresponding to each of the enhanced images and the plurality of channel images respectively; an iteration module for performing iterative processing on the plurality of multi-layer scale images corresponding to each of the enhanced images and the plurality of channel images respectively to obtain a first target image; the a-th iteration in the iterative processing comprises: analyzing a target scale image of the a-th layer, a second scale image, a third scale image and a plurality of enhanced scale images to obtain a correction coefficient of the a+1-th layer; and correcting the first scale image of the a+1-th layer according to the correction coefficient of the a+1-th layer to obtain a target scale image of the a+1-th layer; wherein a is a positive integer, a=1, the target scale image of the a-th layer is the first scale image of the a-th layer, and the first target image is the target scale image of the last layer; the correction coefficient of the a+1-th layer indicates an influence of each enhancement algorithm on each pixel point in the target scale image of the a+1-th layer; an input module for inputting the second target image into a trained neural network model to obtain an analysis result; the second target image is obtained by performing image restoration on the first target image and the second channel image and the third channel image. The target scale image and the first scale image are scale images in the plurality of multi-layer scale images corresponding to the first channel image, the second scale image is a scale image in the plurality of multi-layer scale images corresponding to the second channel image, the third scale image is a scale image in the plurality of multi-layer scale images corresponding to the third channel image, and the plurality of enhanced scale images are scale images in the plurality of multi-layer scale images corresponding to each of the enhanced images respectively. 2.The thoracoscope image-based lung nodule intelligent identification system of claim 1, wherein, The first channel image is a luminance channel image, the second channel image is a hue channel image, and the third channel image is a saturation channel image. 3.The thoracoscope image-based lung nodule intelligent identification system of claim 1, wherein, The analysis of the target scale image of the a-th layer, the second scale image, the third scale image and the plurality of enhanced scale images to obtain the correction coefficient of the a+1-th layer comprises: analyzing a plurality of difference images between the target scale image of the a-th layer and the plurality of enhanced scale images, a pixel value difference of a first target point and a surrounding area of the first target point in the second scale image of the a-th layer, and a pixel value difference of a second target point and a surrounding area of the second target point in the third scale image of the a-th layer to obtain the correction coefficient of the a+1-th layer; wherein the positions of the first target point in the second scale image and the second target point in the third scale image are in one-to-one correspondence. The scale of the a-th layer scale image is greater than the scale of the a+1-th layer scale image in the same multi-layer scale image; the target scale image and the first scale image are scale images in a multi-layer scale image corresponding to a first channel image pair, the second scale image is a scale image in a multi-layer scale image corresponding to a second channel image pair, the third scale image is a scale image in a multi-layer scale image corresponding to a third channel image pair, and the plurality of enhanced scale images are scale images in a multi-layer scale image corresponding to each enhanced image respectively. 4.The thoracoscope image-based lung nodule intelligent identification system of claim 3, wherein, The pixel value difference of the first target point and the surrounding area in the second scale image of the a-th layer and the pixel value difference of the second target point and the surrounding area in the third scale image of the a-th layer are analyzed to obtain the correction coefficient of the a+1-th layer, and the method comprises the following steps: According to the plurality of difference images between the target scale image of the a-th layer and the plurality of enhanced scale images, the plurality of enhancement coefficients of the plurality of enhancement algorithms in the a-th layer are determined; the enhancement coefficient represents the enhancement ability of different enhancement algorithms in different scales; According to the pixel value difference of the first target point and the surrounding area in the second scale image of the a-th layer and the pixel value difference of the second target point and the surrounding area in the third scale image of the a-th layer, the conduction inhibition coefficient of the target point in the target scale image of the a-th layer is determined; wherein the position of the target point in the first scale image corresponds to the position of the first target point in the second scale image one by one; The plurality of enhancement coefficients of the plurality of enhancement algorithms in the a-th layer and the conduction inhibition coefficient of the target point in the target scale image of the a-th layer are analyzed to obtain the correction coefficient of the a+1-th layer. 5.The thoracoscope image-based lung nodule intelligent identification system of claim 4, wherein, The conduction inhibition coefficient of the target point in the target scale image of the a-th layer is determined according to the pixel value difference of the first target point and the surrounding area in the second scale image of the a-th layer and the pixel value difference of the second target point and the surrounding area in the third scale image of the a-th layer, and the method comprises the following steps: According to the absolute difference value between the pixel value of the first target point and the pixel mean value of the first target area in the second scale image of the a-th layer, the first conduction inhibition coefficient of the first target point is determined; the first target area is determined according to the first target point; According to the absolute difference value between the pixel value of the second target point and the pixel mean value of the second target area in the third scale image of the a-th layer, the second conduction inhibition coefficient of the second target point is determined; the second target area is determined according to the second target point; According to the mean value of the first conduction inhibition coefficient and the second conduction inhibition coefficient, the conduction inhibition coefficient of the target point in the target scale image of the a-th layer is determined. 6.The thoracoscope image-based lung nodule intelligent identification system of claim 4, wherein, The plurality of enhancement coefficients of the plurality of enhancement algorithms in the a-th layer and the conduction inhibition coefficient of the target point in the target scale image of the a-th layer are analyzed to obtain the correction coefficient of the a+1-th layer, and the method comprises the following steps: The conduction weight of the target point in the target scale image of the a-th layer is determined according to the plurality of enhancement coefficients of the plurality of enhancement algorithms in the a-th layer, the plurality of dominant coefficients of the plurality of enhancement algorithms in the a-th layer, the conduction inhibition coefficient of the target point in the target scale image of the a-th layer, and the local weight of the target point in the target scale image of the a-th layer of the plurality of enhancement algorithms; wherein the local weight represents the influence of the plurality of enhancement algorithms on the target point in the target scale image of the a-th layer; the a-th layer dominant coefficient of each enhancement algorithm is obtained according to the enhancement coefficient of the corresponding enhancement algorithm in the a-th layer; The correction coefficient of the a+1-th layer is obtained by analyzing the conduction weight of the target point in the target scale image of the a-th layer and the fluctuation weight of the third target point in the first scale image of the a+1-th layer; wherein the fluctuation weight is obtained according to the pixel difference between the target scale image of the a-th layer and the first scale image of the a+1-th layer, and the position of the third target point in the first scale image corresponds to the position of the first target point in the second scale image. 7.The thoracoscope image-based lung nodule intelligent identification system of claim 6, wherein, The determination of the local weight of the target point in the target scale image of the a-th layer of the plurality of enhancement algorithms comprises: when a=1, the local weight of the target point in the target scale image of the a-th layer of the plurality of enhancement algorithms is determined according to the plurality of difference images between the target scale image of the a-th layer and the plurality of enhancement scale images of the a-th layer, and the average pixel value of the target scale image of the a-th layer; when a≥2, the local weight of the target point in the target scale image of the a-th layer of the plurality of enhancement algorithms is determined according to the correction coefficient of the a-1-th layer. 8.The thoracoscope image-based lung nodule intelligent identification system of claim 6, wherein, The analysis of the conduction weight of the target point in the target scale image of the a-th layer and the fluctuation weight of the third target point in the first scale image of the a+1-th layer to obtain the correction coefficient of the a+1-th layer comprises: The correction coefficient of the a+1-th layer is determined according to the sum of the conduction weight of the target point in the target scale image of the a-th layer and the fluctuation weight of the third target point in the first scale image of the a+1-th layer. 9.The thoracoscope image-based lung nodule intelligent identification system of claim 1, wherein, The correction of the first scale image of the a+1-th layer according to the correction coefficient of the a+1-th layer to obtain the target scale image of the a+1-th layer comprises: The target scale image of the a+1-th layer is determined according to the product of the pixel value of each pixel point of the first scale image of the a+1-th layer and the correction coefficient of the a+1-th layer. 10.The thoracoscope image-based lung nodule intelligent identification system of claim 1, wherein, The plurality of enhancement algorithms at least includes the following two kinds: histogram equalization, anisotropic filtering, adaptive histogram truncation, and gamma correction.

Images