Modeling method of hyperspectral rapid detection model for bacterial infection state of urine sample

Abstract

The invention provides a modeling method of a hyperspectral rapid detection model for the bacterial infection state of a urine sample, which is used for solving the technical problems that the bacterial hyperspectral research mostly focuses on the classification and identification of several types of specific bacteria at present, and the application of a urine bacterial infection state model is not available. According to the modeling method of the hyperspectral rapid detection model for the bacterial infection state of the urine sample, the urine hyperspectral data and the novel multi-scale buffer convolutional neural network are combined, more and finer deep map features can be extracted, and rapid screening of abnormal samples is achieved; meanwhile, the microscopic hyperspectral technology is easy and convenient to operate, a large amount of culture time is saved, the overall detection time is greatly shortened, and the detection cost is reduced.

Description

Technical Field

[0001] The invention relates to a modeling method for a urine sample bacterial infection state model, and in particular to a modeling method for a high-spectrum rapid detection model for a urine sample bacterial infection state. Background Art

[0002] At present, the current urine sample bacterial infection detection technology in clinical practice requires blood culture instrument enrichment culture, positive transfer blood plate, separation culture identification and other processes, and the overall detection efficiency is low. The most time-consuming bacterial culture process takes 1-2 days, and the concentration of bacterial suspension is usually >10 5 CFU / ml, positive, concentration <10 3 In the case of CFU / ml, it is directly assumed to be negative. The current detection technology has the disadvantages of long bacterial culture cycle and insufficient detection throughput, which makes it difficult to meet the clinical needs of rapid detection.

[0003] Currently, most bacterial hyperspectral research focuses on the classification and identification of several specific types of bacteria. For example, Matthew used visible light / near-infrared hyperspectral microscopy to image salmonella in chicken rinse fluid for detection. For the 100× magnification of salmonella colony hyperspectral data at 450-800nm, the use of secondary discriminant analysis achieved a classification accuracy of 98.5% and a specificity of 0.963. Liu combined hyperspectral microscopy with machine learning to achieve a classification accuracy of 98.06% for two types of bacteria, B.megaterium and B.cereus, based on the subtle differences in absorption peaks. However, the application of hyperspectral technology to urine bacterial infection status models has not yet been seen. Summary of the invention

[0004] The purpose of the present invention is to solve the technical problem that most of the current bacterial hyperspectral research is focused on the classification and identification of several specific types of bacteria, and there is no application of urine bacterial infection status model, and to provide a modeling method for a hyperspectral rapid detection model of bacterial infection status of urine samples.

[0005] In order to achieve the above object, the technical solution provided by the present invention is as follows:

[0006] A method for modeling a hyperspectral rapid detection model for bacterial infection status of urine samples is special in that it includes the following steps:

[0007] Step 1, smearing a urine sample to obtain a urine smear; performing hyperspectral acquisition on a field of view of suspected bacterial distribution in the urine smear to obtain a hyperspectral image of the urine smear;

[0008] Step 2, establishing a multi-scale buffered convolutional neural network;

[0009] The multi-scale buffered convolutional neural network includes a convolution combination unit, a feature splicing layer, four buffer units, a fully connected layer and an activation function layer arranged in sequence according to input and output; the convolution combination unit includes at least three convolution layers arranged in parallel and with different convolution kernel sizes, one of which is used to obtain the spectral information characteristics of the urine smear hyperspectral image, and the remaining convolution layers are used to obtain spatial texture information characteristics of different scales of the urine smear hyperspectral image; the feature splicing layer is used to splice the feature map output by each convolution layer; the buffer unit includes three convolution layers arranged in sequence according to input and output, and the four buffer units are used to further extract features from the spliced ​​feature map in sequence; the fully connected layer is used to receive the features extracted by the fourth buffer unit and make classification decisions; the activation function layer is used to output prediction results according to the classification decisions;

[0010] Step 3, input the urine smear hyperspectral image obtained in step 1 into the multi-scale buffered convolutional neural network for training. The trained multi-scale buffered convolutional neural network is a hyperspectral rapid detection model for the bacterial infection status of urine samples.

[0011] Furthermore, the method further comprises step 4:

[0012] Establishing a hyperspectral database of urine bacteria and urine impurities, the hyperspectral database includes standard strains, genus and species information of clinical samples, microscopic data cubes, standard spectral curves, typical spectral bands, visualized spectral features, and medical examination parameters;

[0013] After the trained multi-scale buffered convolutional neural network performs probability prediction on the bacterial infection status of the urine smear hyperspectral image to be tested, the hyperspectral database is applied to the multi-scale buffered convolutional neural network to form a hyperspectral rapid detection model for the bacterial infection status of urine samples;

[0014] The application of the hyperspectral database to the multi-scale buffered convolutional neural network specifically includes: extracting the probability of the multi-scale buffered convolutional neural network output to predict all single-target samples of the normal urine smear hyperspectral image to be tested, and matching the spectral curves of all single-target samples with the hyperspectral database of urine bacteria and urine impurities.

[0015] Furthermore, in step 4, the hyperspectral database of urine bacteria and urine impurities is established by: determining the true value of the positive and negative bacterial infection of the urine sample after traditional culture, staining, biochemical molecular diagnosis and mass spectrometry identification; and establishing a hyperspectral database of urine bacteria and urine impurities based on the determined true value of the positive and negative bacterial infection.

[0016] Further, in step 2, the number of convolutional layers in the convolutional combination unit is three, and the corresponding convolutional kernel sizes are 3×1×1, 3×3×3, and 3×5×5, respectively, wherein the convolutional layer with a convolutional kernel size of 3×1×1 is used to obtain the spectral information characteristics of the urine smear hyperspectral image, and the convolutional layers with convolutional kernel sizes of 3×3×3 and 3×5×5 are used to obtain spatial texture features of different scales of the urine smear hyperspectral image;

[0017] The convolution kernel size of each convolution layer in each of the buffer units is 3×3×3;

[0018] The step size of the convolution layer in the convolution combination unit is set to 2, and the step size of the convolution layer in each buffer unit is set to 1.

[0019] Furthermore, in step 2, the convolution combination unit further includes three pooling layers, and the input ends of the three pooling layers are connected to the output ends of the convolutional layers; the input end of the feature concatenation layer is respectively connected to the output ends of the three pooling layers;

[0020] Each of the buffer units also includes a pooling layer connected to the output end of each convolutional layer;

[0021] The input end of the fully connected layer is connected to the output end of the pooling layer in the fourth buffer unit;

[0022] The step size of each pooling layer in the convolution combination unit is set to 2, and the step size of the pooling layer in each buffer unit is set to 2.

[0023] Furthermore, in step 4, the hyperspectral database of urine bacteria and urine impurities is established as follows:

[0024] After traditional culture, staining, biochemical molecular diagnosis and mass spectrometry identification of urine samples, the true value of the positive and negative bacterial infection is determined; based on the determined true value of the positive and negative bacterial infection, a high-spectral database of urine bacteria and urine impurities is established.

[0025] Furthermore, in step 1, the urine sample of the experimental group is smeared, and the urine smear is obtained specifically as follows:

[0026] Step a1, take a clean glass slide and pre-treat it;

[0027] Step b1, pour the urine sample of the experimental group into an anticoagulant tube, balance it and place it in a centrifuge for centrifugation; after the centrifugation is completed, use a clean and sterile pipette to suck out the supernatant, leaving the urine sediment at the bottom;

[0028] Step c1, suck out the urine sediment with a pipette, blow and mix it, and then evenly apply it on the pretreated slide;

[0029] Step d1, placing the slide coated with urine sediment in a biosafety cabinet and waiting for it to completely dry, then staining it with crystal violet stain, covering it with iodine solution, decolorizing it with 95% ethanol, re-staining it with safranin stain, and finally washing it with water and drying it to obtain a urine smear.

[0030] Furthermore, in step 1, the field of view of suspected bacterial distribution in the urine smear is collected by hyperspectral acquisition to obtain the urine smear hyperspectral image specifically as follows:

[0031] Step a2, placing the obtained urine smear on the microscope stage, first looking for the overall field of view under a 10x objective lens, and then looking for the field of view of suspected bacterial distribution under a 100x objective lens;

[0032] Step b2, using a halogen lamp to provide an active lighting source, and using a hyperspectral imaging system to perform hyperspectral acquisition on the field of view of suspected bacterial distribution in the urine smear to obtain a hyperspectral image of the urine smear.

[0033] Furthermore, step a1 is specifically as follows:

[0034] Take a clean glass slide, disinfect it with alcohol and rinse it with distilled water, then use an alcohol lamp to bake and remove wax. After cooling, the pretreatment is completed.

[0035] Furthermore, step a1 is specifically as follows:

[0036] In step b1, the rotation speed of the centrifuge is 3500r / 10min;

[0037] In step c1, a sterile inoculating loop is used to quickly and evenly apply the mixed urine sediment onto the pretreated glass slide.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. The modeling method of the hyperspectral rapid detection model for bacterial infection status of urine samples provided by the present invention, combined with urine hyperspectral data and a new multi-scale buffered convolutional neural network, can extract more and more refined deep spectral features and realize rapid screening of abnormal samples; at the same time, the microscopic hyperspectral technology is easy to operate, saves a lot of culture time, greatly shortens the overall detection time, and reduces the detection cost.

[0040] 2. Based on the application of hyperspectral images to multi-scale buffered convolutional neural networks, the present invention combines single target extraction with spectral angle matching to make up for the defect of low sensitivity of multi-scale buffered convolutional neural networks to small targets.

[0041] 3. The modeling method of the hyperspectral rapid detection model for the bacterial infection status of urine samples provided by the present invention can encode the biological characteristics of bacteria into three-dimensional data through spectral and morphological information representation at a microscopic scale, and can extract more and more refined deep map features through appropriate preprocessing methods and deep learning models, so that the overall process does not rely on traditional morphological observations and can reduce the influence of human factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A flow chart of an embodiment of a modeling method of a hyperspectral rapid detection model for bacterial infection status of urine samples according to the present invention;

[0043] Figure 2 Schematic diagram of the configuration of the multi-scale buffered convolutional neural network in step 2 of an embodiment of the present invention;

[0044] Figure 3 Schematic diagram of the single target data cube and its pseudo-color image in step 4.2 of an embodiment of the present invention, wherein (a) is a pseudo-color image of Candida tropicalis and its corresponding single target data cube, and (b) is a pseudo-color image of a magnesium ammonium phosphate sample and its corresponding single target data cube. DETAILED DESCRIPTION

[0045] In order to make the advantages and features of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0046] like Figure 1 A modeling method for a high-spectrum rapid detection model of bacterial infection status of a urine sample is shown, which specifically includes the following steps:

[0047] Step 1: Quickly obtain the urine sample data to be tested.

[0048] Step 1.1, set aside a portion of the urine sample to be tested as the experimental group, and the other portion as the control group.

[0049] Step 1.2, smear the urine sample of the experimental group to obtain a urine smear. The specific process is as follows: take a clean glass slide, first disinfect the glass slide with alcohol and rinse it with distilled water, then use an alcohol lamp to bake it to remove wax, and cool it for use. After recording the detailed information of the urine sample to be tested and numbering it, draw the urine into the anticoagulant tube and balance it; after balancing, place the anticoagulant tube in a centrifuge and centrifuge it at a speed of 3500r / 10min. Take out the anticoagulant tube that has completed the centrifugation operation, use a clean and sterile pipette to suck out the supernatant, leaving the urine sediment at the bottom, then use a pipette to suck out the urine sediment and blow and mix it; use a sterile inoculation loop to quickly and evenly apply the blown and mixed urine sediment on the pretreated glass slide. The slide coated with urine sediment was placed in a biosafety cabinet and allowed to dry completely. It was then stained with crystal violet solution, covered with iodine solution, decolorized with 95% ethanol, restained with safranin solution, and finally washed with water and dried to obtain a urine smear.

[0050] Step 1.3, place the obtained urine smear on the microscope stage, first find the overall field of view under the 10x objective lens, then convert the microscope to the 100x objective lens, and then find the field of view of suspected bacterial distribution under the 100x objective lens. After finding the field of view of suspected bacterial distribution, a halogen lamp is used to provide an active lighting source, and a hyperspectral imaging system is used to perform hyperspectral acquisition of the field of view of suspected bacterial distribution in the urine smear on the microscope stage to obtain a hyperspectral image of the urine smear. The halogen lamp used in this embodiment has a wavelength range of 400nm to 2500nm and a power of 50W.

[0051] Step 1.4: After traditional culture, staining, biochemical molecular diagnosis and mass spectrometry identification, the urine samples of the control group were used to determine the true value of bacterial infection and mark the corresponding urine sample number. The types of bacterial infection include Escherichia coli, Klebsiella pneumoniae, Acinetobacter baumannii, Proteus mirabilis, Enterococcus faecium, Staphylococcus epidermidis, Pseudomonas aeruginosa, Staphylococcus aureus, Candida albicans and Candida tropicalis.

[0052] Step 2: Establish a multi-scale buffered convolutional neural network.

[0053] In the field of medical imaging, most end-to-end models based on three-dimensional convolutional networks are proposed to process stereoscopic imaging modes such as CT and MRI. Three-dimensional convolutional networks have more parameters, higher complexity, and are more difficult to train, which is inconsistent with the small sample size of urine bacterial hyperspectral data. To this end, in view of the characteristics of urine bacterial data, the present invention proposes a multi-scale buffered convolutional neural network, in which multi-scale refers to the use of convolution kernels of different sizes to process the feature map of the same layer, combining different convolution kernels in parallel, and merging the convolution results. The use of convolution kernels of different sizes is to introduce receptive fields of different sizes and extract features of different scales. In view of the characteristics of rich spectral information and differences in target size in urine bacterial data, a convolution combination unit containing 3 groups of convolution kernels is designed in this embodiment.

[0054] like Figure 2As shown, the multi-scale buffered convolutional neural network of the present invention includes a convolutional combination unit, a feature splicing layer, four buffer units, a fully connected layer, and an activation function layer. In view of the characteristics of rich spectral information and target size differences in urine bacterial data, the convolutional combination unit includes three parallel convolutional layers with different convolution kernel sizes and three pooling layers arranged after each convolutional layer. The input ends of the three convolutional layers are all used as input ports of the multi-scale buffered convolutional neural network to receive external input urine smear hyperspectral images respectively. The convolution kernel sizes of the three convolution layers in the convolution combination unit of this embodiment are 3×1×1 (spectral dimension×number of rows of spatial dimension×number of columns of spatial dimension), 3×3×3, and 3×5×5, respectively, wherein the convolution layer with a convolution kernel size of 3×1×1 is used to obtain the spectral information characteristics of the urine smear hyperspectral image, and the convolution layers with convolution kernel sizes of 3×3×3 and 3×5×5 are used to obtain spatial texture features of different scales of the urine smear hyperspectral image; the three pooling layers in the convolution combination unit are respectively used to downsample the features extracted by the corresponding convolution layers to compress the data dimension, the number of parameters, and the amount of calculation. The input end of the feature splicing layer is respectively connected to the output end of the three pooling layers in the convolution combination unit, and is used to splice the feature maps processed by the three pooling layers. The buffer unit includes three convolution layers and a pooling layer arranged in sequence according to input and output, wherein the input end of the first convolution layer in the first buffer unit is connected to the output end of the feature splicing layer, and the input end of the first convolution layer in the second to fourth buffer units is respectively connected to the output end of the pooling layer in the previous buffer unit. The convolution kernel size of each convolution layer in the buffer unit is 3×3×3, which is used to extract detailed information of the data cube; the role of the pooling layer in the buffer unit is similar to that of the pooling layer in the convolution combination unit. The step size of each convolution layer in the convolution combination unit is set to 2, the step size of the pooling layer is set to 2, the step size of the convolution layer in each buffer unit is set to 1, and the step size of the pooling layer is set to 2 to enhance the representation ability without reducing the resolution of the feature map. The input end of the fully connected layer is connected to the output end of the pooling layer in the fourth buffer unit to receive the features extracted by the fourth buffer unit and make classification decisions; the activation function layer is used to output the probability prediction results based on the classification decision.

[0055] Step 3: Input the urine smear hyperspectral image obtained from the experimental group into the multi-scale buffered convolutional neural network for training to obtain a trained multi-scale buffered convolutional neural network.

[0056] The collected urine smear hyperspectral image to be tested is input into the trained multi-scale buffered convolutional neural network, and the urine smear hyperspectral image to be tested is also obtained by the method of steps 1.2 to 1.3. In this embodiment, the size of the urine smear hyperspectral image is 160×400×400, and the three convolutional layers in the convolutional combination unit respectively output 6 feature maps of size 78×198×198, and after pooling by the corresponding pooling layer, the output size is 39×99×99 feature maps, and then 18 39×99×99 feature maps are input to the feature splicing layer for splicing. The spliced ​​feature maps enter the four buffer combination units in turn, and the feature map sizes output by the four buffer combination units are 20×50×50, 10×25×25, 5×13×13, and 3×7×7, respectively, and the number of output feature maps is 32, 64, 128, and 256, respectively, that is, the fourth buffer combination unit finally outputs 256 feature maps of size 3×7×7. Then, the probability value is output through the fully connected layer and the activation function layer. If the probability value indicates that the urine sample to be tested is positive, the result is output as the final result to complete the infection status determination; if the probability value indicates that the urine sample to be tested is negative, the next step of determination is performed through step 4.

[0057] Step 4: For the multi-scale buffered convolutional neural network, its model architecture focuses more on global or large-scale features, and may ignore single bacterial targets that appear sporadically in some areas. Spectral database matching focuses more on the characteristics of small targets. Therefore, the present invention adopts a single target extraction combined with spectral angle matching to improve the sensitivity of the multi-scale buffered convolutional neural network to small targets.

[0058] Step 4.1, establish a hyperspectral database of urine bacteria and urine impurities in combination with the true value of the positive and negative bacterial infection determined by the control group. On the one hand, the hyperspectral database needs to include standard strains, genus and species information of clinical samples, microscopic data cubes, standard spectral curves, typical spectral bands, visual spectrum features, medical examination parameters, and basic patient information, among which standard strains include Candida tropicalis, Staphylococcus aureus, Klebsiella pneumoniae, Staphylococcus epidermidis, Escherichia coli, Pseudomonas aeruginosa and other bacterial spectral data. On the other hand, in view of the interference of urine impurities (etc.), urine impurity spectral data including urine cells, casts, and crystals are also required, such as calcium oxalate crystals, magnesium ammonium phosphate crystals, ammonium urate crystals, lamellar calcium phosphate crystals, casts, leucine crystals and other impurity spectral data.

[0059] Step 4.2, in order to determine whether the urine sample to be tested has bacterial infection, that is, whether there are bacteria in the urine sample to be tested, it is also necessary to obtain a single target (single bacteria) hyperspectral data cube. In this embodiment, the K-means clustering algorithm is used to separate the foreground target and the background area of ​​the urine smear hyperspectral image to obtain a binary image of the target and the background. Then, combined with the bacterial morphological parameters, impurities and unknown bacteria with abnormal parameters are removed in the foreground, and the incomplete bacteria and incomplete impurities on the image boundary are excluded. The remaining single target can be retained as a single target hyperspectral data cube, that is, a single target sample. Figure 3 As shown, (a) is a pseudo-color image of Candida tropicalis and its corresponding single target data cube, and (b) is a pseudo-color image of magnesium ammonium phosphate sample and its corresponding single target data cube. The red box is the location of the single target, and the small picture in the lower right corner is the corresponding single target data cube.

[0060] Preferably, by setting several constraints such as the single target size range, aspect ratio range, and single connected area restriction, single target samples can be automatically extracted. The number of spectral segments of the obtained single target samples is the same, but the sample space dimensions are different due to differences in their own spatial sizes.

[0061] Step 4.3, extract all single-target samples in the urine smear hyperspectral image, and then obtain the spectral curves of all single-target samples to match them with the hyperspectral database of urine bacteria and urine impurities; if the spectral angle between any single-target sample and a substance in the hyperspectral database is lower than the set threshold, the current single-target sample is judged to be the substance, if the substance belongs to a known bacterial species, it is judged to be bacterial, if the substance belongs to an impurity or others, it is judged to be sterile; if the spectral angle between all single-target samples and all substances in the hyperspectral database is higher than the threshold, it is judged to be sterile; if a bacterial judgment occurs during the matching process, it is directly output as positive, if no bacterial judgment occurs, it is output as negative.

[0062] After matching the spectral database, single bacterial targets that were previously ignored by the multi-scale buffered convolutional neural network were discovered, and thus a large number of positive samples that were misjudged as negative were corrected.

[0063] The above description is only used to illustrate the technical solution of the present invention rather than to limit it. For ordinary professional and technical personnel in the field, the specific technical solution recorded in the above embodiment can be modified, or some of the technical features therein can be replaced by equivalents, and these modifications or replacements do not make the essence of the corresponding technical solution deviate from the scope of the technical solution protected by the present invention.

Claims

1. A method for modeling a hyperspectral rapid detection model for bacterial infection status of urine samples, characterized in that: The following steps are involved: Step 1, smearing a urine sample to obtain a urine smear; performing hyperspectral acquisition on a field of view of suspected bacterial distribution in the urine smear to obtain a hyperspectral image of the urine smear; Step 2, establishing a multi-scale buffered convolutional neural network; The multi-scale buffered convolutional neural network includes a convolutional combination unit, a feature splicing layer, four buffer units, a fully connected layer, and an activation function layer arranged in sequence according to input and output; the convolutional combination unit includes at least three convolutional layers arranged in parallel and having different convolution kernel sizes, one of which is used to obtain the spectral information characteristics of the urine smear hyperspectral image, and the remaining convolutional layers are used to obtain spatial texture information characteristics of different scales of the urine smear hyperspectral image; The feature concatenation layer is used to concatenate the feature maps output by each convolution layer; the buffer unit includes three convolution layers arranged in sequence according to input and output, and the four buffer units are used to further extract features from the concatenated feature maps in sequence; The fully connected layer is used to receive the features extracted by the fourth buffer unit and make classification decisions; the activation function layer is used to output the prediction results based on the classification decision; Step 3, input the urine smear hyperspectral image obtained in step 1 into the multi-scale buffered convolutional neural network for training. The trained multi-scale buffered convolutional neural network is a hyperspectral rapid detection model for the bacterial infection status of urine samples.

2. The modeling method of the hyperspectral rapid detection model of the bacterial infection status of urine samples according to claim 1 is characterized in that: Also includes step 4: Establishing a hyperspectral database of urine bacteria and urine impurities, the hyperspectral database includes standard strains, genus and species information of clinical samples, microscopic data cubes, standard spectral curves, typical spectral bands, visualized spectral features, and medical examination parameters; After the trained multi-scale buffered convolutional neural network performs probability prediction on the bacterial infection status of the urine smear hyperspectral image to be tested, the hyperspectral database is applied to the multi-scale buffered convolutional neural network to form a hyperspectral rapid detection model for the bacterial infection status of urine samples; The application of the hyperspectral database to the multi-scale buffered convolutional neural network specifically includes: extracting the probability of the multi-scale buffered convolutional neural network output to predict all single-target samples of the normal urine smear hyperspectral image to be tested, and matching the spectral curves of all single-target samples with the hyperspectral database of urine bacteria and urine impurities.

3. The modeling method of the hyperspectral rapid detection model of the bacterial infection status of urine samples according to claim 2 is characterized by: In step 4, the hyperspectral database of urine bacteria and urine impurities is established as follows: After traditional culture, staining, biochemical molecular diagnosis and mass spectrometry identification of urine samples, the true value of the positive and negative bacterial infection is determined; based on the determined true value of the positive and negative bacterial infection, a high-spectral database of urine bacteria and urine impurities is established.

4. The modeling method of the hyperspectral rapid detection model of the bacterial infection status of urine samples according to any one of claims 1 to 3, characterized in that: In step 2, the number of convolutional layers in the convolutional combination unit is three, and the corresponding convolutional kernel sizes are 3×1×1, 3×3×3, and 3×5×5, respectively, wherein the convolutional layer with a convolutional kernel size of 3×1×1 is used to obtain the spectral information characteristics of the urine smear hyperspectral image, and the convolutional layers with convolutional kernel sizes of 3×3×3 and 3×5×5 are used to obtain spatial texture features of different scales of the urine smear hyperspectral image; The convolution kernel size of each convolution layer in each of the buffer units is 3×3×3; The step size of the convolution layer in the convolution combination unit is set to 2, and the step size of the convolution layer in each buffer unit is set to 1.

5. The method for building a model for a rapid hyperspectral detection model of bacterial infection status in urine samples according to claim 4, characterized in that: In step 2, the convolution combination unit further includes three pooling layers, the input ends of the three pooling layers are connected to the output ends of the convolution layers; the input end of the feature concatenation layer is respectively connected to the output ends of the three pooling layers; Each of the buffer units also includes a pooling layer connected to the output end of each convolutional layer; The input end of the fully connected layer is connected to the output end of the pooling layer in the fourth buffer unit; The step size of each pooling layer in the convolution combination unit is set to 2, and the step size of the pooling layer in each buffer unit is set to 2.

6. The modeling method of the hyperspectral rapid detection model of the bacterial infection status of urine samples according to claim 1, characterized in that: In step 1, the urine sample of the experimental group is smeared, and the urine smear is obtained as follows: Step a1, take a clean glass slide and pre-treat it; Step b1, pour the urine sample of the experimental group into an anticoagulant tube, balance it and place it in a centrifuge for centrifugation; after the centrifugation is completed, use a clean and sterile pipette to suck out the supernatant, leaving the urine sediment at the bottom; Step c1, suck out the urine sediment with a pipette, blow and mix it, and then evenly apply it on the pretreated slide; Step d1, placing the slide coated with urine sediment in a biosafety cabinet and waiting for it to completely dry, then staining it with crystal violet stain, covering it with iodine solution, decolorizing it with 95% ethanol, re-staining it with safranin stain, and finally washing it with water and drying it to obtain a urine smear.

7. The method for building a model for a rapid hyperspectral detection model of bacterial infection status in urine samples according to claim 6, characterized in that: In step 1, a hyperspectral image is acquired for the field of view of suspected bacterial distribution in the urine smear to obtain a urine smear hyperspectral image as follows: Step a2, placing the obtained urine smear on the microscope stage, first looking for the overall field of view under a 10x objective lens, and then looking for the field of view of suspected bacterial distribution under a 100x objective lens; Step b2, using a halogen lamp to provide an active lighting source, and using a hyperspectral imaging system to perform hyperspectral acquisition on the field of view of suspected bacterial distribution in the urine smear to obtain a hyperspectral image of the urine smear.

8. The modeling method of the hyperspectral rapid detection model of the bacterial infection status of urine samples according to claim 7 is characterized in that: Step a1 is specifically as follows: Take a clean glass slide, disinfect it with alcohol and rinse it with distilled water, then use an alcohol lamp to bake and remove wax. After cooling, the pretreatment is completed.

9. The modeling method of the hyperspectral rapid detection model of the bacterial infection status of urine samples according to claim 8, characterized in that: Step a1 is specifically as follows: In step b1, the rotation speed of the centrifuge is 3500r / 10min; In step c1, a sterile inoculating loop is used to quickly and evenly apply the mixed urine sediment onto the pretreated glass slide.

Images