Medical sterilizer sterilization time compensation method and system

Abstract

The invention discloses a medical sterilizer sterilization time compensation method and system, and the method comprises the steps: S1, collecting and preprocessing an original sterilization image, and obtaining a preprocessed sterilization image; s2, according to the preprocessed sterilization image, calculating to obtain a liquid region segmentation image; then morphological opening operation is carried out, and finally feature extraction is carried out to obtain sterilization image features; s3, liquid target identification is carried out to obtain a liquid target frame set, then liquid target frame elimination is carried out, and finally, binarization processing is carried out on the preprocessed sterilization image to obtain a binarized sterilization image; s4, counting the number of pixels in a white area of the binarized sterilization image, and calculating the liquid volume and the liquid weight; and S5, acquiring the initial sterilization time of the to-be-sterilized object, compensating the sterilization time, and sterilizing the to-be-sterilized object. The medical sterilizer can solve the problem that a traditional medical sterilizer is poor in sterilization effect due to the fact that thermal inertia of liquid is not fully considered.

Description

Technical Field

[0001] The present invention relates to the field of medical sterilizers, and in particular to a sterilization time compensation method and system for a medical sterilizer. Background Art

[0002] In the medical field, sterilization of medical items containing liquids is very important and directly related to medical safety. However, due to the large thermal inertia of liquids, the response to temperature changes is relatively slow, resulting in the microorganisms inside the liquid not being fully killed during the sterilization process, thus affecting the sterilization effect. Some experienced operators may appropriately extend the sterilization time based on experience according to the type and quantity of the liquid, but this approach lacks scientificity and accuracy, and often cannot ensure the reliability of the sterilization effect. Therefore, there is an urgent need for a scientific and accurate time compensation method to accurately calculate the compensation time according to the characteristics of the liquid, thereby improving the sterilization effect and ensuring medical safety. Summary of the invention

[0003] In view of this, the present invention provides a sterilization time compensation method and system for a medical sterilizer, aiming to solve the problem of poor sterilization effect caused by insufficient consideration of liquid thermal inertia in traditional medical sterilizers.

[0004] A method for compensating sterilization time of a medical sterilizer, comprising:

[0005] S1: collecting the original sterilization image and preprocessing it to obtain the preprocessed sterilization image;

[0006] S2: Based on the pre-processed sterilization image, the liquid area segmentation image is calculated; then a morphological opening operation is performed to obtain a segmentation image after the opening operation; finally, feature extraction is performed on the pre-processed sterilization image to obtain sterilization image features, including:

[0007] According to the preprocessed sterilization image, an improved U-Net network is used to perform semantic segmentation; the improved U-Net network, in the downsampling path, designs an adaptive feature fusion module, combines a multi-layer perceptron to calculate the fusion coefficient, and adaptively fuses the feature map after convolution and the feature map after pooling;

[0008] Convolution kernel sizes and convolution steps of various scales are used to extract the sterilization image feature maps of each branch. Then, a convolutional neural network with a convolution kernel size of 1×1 is used to extract preliminary sterilization image features. Finally, the attention weights of the sterilization image feature maps of each branch are calculated using the attention mechanism, and weighted summation is performed to obtain the sterilization image features.

[0009] S3: According to the pre-processed sterilization image, the segmented image after the opening operation and the sterilization image features, liquid target recognition is performed to obtain a liquid target frame set; then, the liquid target frame is eliminated according to the sterilization image features to obtain a liquid target frame set after elimination; finally, according to the liquid target frame set after elimination, the pre-processed sterilization image is binarized to obtain a binarized sterilization image;

[0010] S4: for the sterilized image after binarization, count the number of pixels in the white area, and calculate the liquid volume and liquid weight;

[0011] S5: Obtain the initial sterilization time of the object to be sterilized, and perform sterilization time compensation and sterilization of the object to be sterilized based on the liquid weight, liquid volume and the binarized sterilization image.

[0012] S1: collecting the original sterilization image and preprocessing it to obtain the preprocessed sterilization image;

[0013] Furthermore, the step S1 further includes:

[0014] S11: Use a high-resolution camera at a fixed height to photograph the object to be sterilized from a top-down perspective to obtain an original sterilization image;

[0015] S12: preprocessing the original sterilization image, including: using an adaptive median filtering algorithm to perform image denoising, using a histogram equalization to perform image enhancement, and obtaining a sterilization image after preliminary preprocessing;

[0016] S13: geometrically correcting the sterilization image after preliminary preprocessing by using affine transformation to obtain a preprocessed sterilization image.

[0017] Furthermore, the step S2 further includes:

[0018] S21: Based on the preprocessed sterilization image, an improved U-Net network is used for semantic segmentation to obtain a liquid area segmentation image, including:

[0019] Firstly, the preprocessed sterilization image is input into the improved U-Net network, which includes a downsampling path and an upsampling path. The upsampling path is used to obtain an upsampling feature map, and the downsampling path is used to obtain a downsampling fused feature map.

[0020] The improved U-Net network designs an adaptive feature fusion module in the downsampling path to adaptively fuse the feature map after convolution and the feature map after pooling. The calculation method is:

[0021]

[0022] in, is the feature map after downsampling and fusion of the Lth layer, L is the layer index, is the feature map after the Lth layer convolution, is the feature map after the Lth layer pooling, α (L) is the fusion coefficient of the Lth layer, sigmoid is the sigmoid function, MLP() is the multi-layer perceptron, and Concat() is the concatenation operation;

[0023] Then, the up-sampled feature map is concatenated with the down-sampled fused feature map, and the feature fusion is performed through a convolutional neural network to finally obtain the liquid area segmentation image. The calculation method is:

[0024]

[0025] Among them, P segmentation is the liquid area segmentation image, CNN ( ) is the convolutional neural network, F up is the upsampled feature map;

[0026] S22: Using morphological opening operation, the liquid area segmentation image is first corroded and then expanded to obtain the segmentation image after opening operation, which is recorded as P' segmentation ;

[0027] S23: Extract features from the pre-processed sterilization image to obtain sterilization image features. The calculation method is:

[0028] F i =CNN(P,k i ,s i );

[0029] F feature =CNN 1×1 (Concat(F1,F2,…,F m ));

[0030] att i =Attention(F i );

[0031]

[0032] Among them, F i is the sterilization image feature map of the i-th branch, i is the branch index, P is the preprocessed sterilization image, k i is the convolution kernel size of the i-th branch, s i is the convolution step size of the i-th branch, F feature For the preliminary sterilization image features, CNN 1×1 is a convolutional neural network with a convolution kernel size of 1×1, F1, F2, …, F mare the sterilization image feature maps of the 1st, 2nd, ..., mth branches, respectively, m is the total number of branches, att i is the attention weight of the sterilization image feature map of the i-th branch, Attention( ) is the attention mechanism, F' feature It is the sterilization image feature.

[0033] It should be noted that the U-Net network is a paper published by Ronneberger et al. in MICCAI, a top journal of computer medical imaging, in 2015. The paper first proposed a U-shaped structure for semantic segmentation of images.

[0034] The present invention improves the U-Net network and designs an adaptive feature fusion module in the downsampling path. The module calculates the fusion coefficient by combining the feature map after convolution and the feature map after pooling through a multi-layer perceptron. The adaptive feature fusion module can dynamically adjust the fusion ratio of convolution features and pooling features, so that the network can better adapt to liquid areas of different scales and shapes. At the same time, the fusion of upsampling and downsampling features further improves the precision and accuracy of segmentation.

[0035] In comparison, the traditional U-Net network only extracts features through convolution and pooling operations in the downsampling path, and restores the feature map size through deconvolution or upsampling operations in the upsampling path. Although this structure is simple, it is difficult to fully capture the association and difference between features when processing complex liquid areas, resulting in inaccurate segmentation results or missing important information, which in turn affects the accuracy of subsequent steps such as liquid target recognition, liquid volume and weight estimation.

[0036] In addition, the present invention adopts a multi-branch convolutional neural network structure, each branch uses a convolution kernel and step size of different sizes to extract features from the preprocessed sterilization image, which can capture features of different scales in the image and improve the diversity and richness of features; after feature extraction, a 1×1 convolutional neural network is used to fuse the feature maps of multiple branches. The 1×1 convolution kernel can adjust the number of channels of the feature map without changing the spatial size of the feature map, thereby realizing linear combination and dimensionality reduction of features; finally, an attention mechanism is introduced to weight the feature map of each branch, so that the model pays more attention to features that are beneficial to the task and suppresses irrelevant or noise features;

[0037] Compared with traditional neural networks, traditional methods usually use single-scale convolutional neural networks for feature extraction, which makes it difficult to capture features of different scales in the image. The sterilization image features extracted by the present invention are diverse and rich, and more accurately represent the liquid information in the image, which helps to more comprehensively analyze and process image data and improve the performance of the entire system.

[0038] Furthermore, the step S3 further includes:

[0039] S31: Liquid target recognition is performed based on the pre-processed sterilization image, the segmented image after the open operation, and the sterilization image features. The calculation method is:

[0040] B j =YOLOv8(ChannelConcat(P,P′ segmentation ,F′ feature ));

[0041] Among them, B j is the jth liquid target frame, j is the liquid target frame index, YOLOv8( ) is the target detection network, ChannelConcat( ) is the channel concatenation; B is calculated j Finally, all liquid target frames are uniformly stored to obtain a liquid target frame set;

[0042] S32: Eliminate the liquid target frame according to the sterilization image features to obtain a set of liquid target frames after elimination. The calculation process is as follows:

[0043] Firstly, in the preprocessed sterilization image, the image areas corresponding to all liquid target frames in the liquid target frame set are cropped, and the cropped areas are pixel-filled using the mean method to obtain the sterilization image background data;

[0044] Secondly, feature extraction is performed on the cropped area and sterilization image background data respectively, and the calculation method is as follows:

[0045]

[0046] f background =CNN(P background );

[0047] in, is the image region feature corresponding to the jth liquid target frame, LBP( ) is the local binary pattern, is the image area corresponding to the jth liquid target frame, CLCM( ) is the gray-level co-occurrence matrix, f background It is the background feature of sterilization image;

[0048] Finally, according to the background features of the sterilization image and the image area features corresponding to the liquid target frame, the liquid target frame rejection threshold and the target frame image area threshold are calculated respectively, and the liquid target frame is rejected. The calculation method is:

[0049] r=sigmoid(W1·f background +b1);

[0050]

[0051] Among them, r is the liquid target frame rejection threshold, W1 is the liquid target frame rejection threshold matrix, b1 is the liquid target frame rejection threshold offset vector, r j is the target frame image area threshold corresponding to the j-th liquid target frame;

[0052] Calculate r and r j After that, if r j >r, the liquid target frame is retained. If r j ≤r, the liquid target frame is removed;

[0053] After traversing and removing all liquid target frames, the remaining liquid target frames are uniformly stored to obtain a set of liquid target frames after removal;

[0054] S33: According to the eliminated liquid target frame set, the pre-processed sterilization image is binarized, the image area corresponding to the liquid target frame is filled with white, and the remaining area is filled with black, to obtain a binarized sterilization image, which is recorded as P binary .

[0055] It should be noted that the present invention performs feature extraction on the cropped liquid target frame area and the sterilization image background area respectively; uses LBP (local binary pattern) and GLCM (gray level co-occurrence matrix) to extract the texture features of the liquid target frame area, LBP generates a binary pattern by comparing the gray values ​​of a pixel with its neighboring pixels to capture local texture information; GLCM analyzes the gray value and spatial relationship of pixel pairs to quantify texture information; then, the liquid target frame rejection threshold and the target frame image area threshold are calculated respectively through linear transformation and sigmoid function;

[0056] Compared with traditional methods, most traditional methods only perform target frame removal based on pixel values ​​or simple morphological operations, which are easily affected by noise and complex background, resulting in false removal or missed removal; after removing inaccurate liquid target frames, the remaining liquid target frames more accurately represent the actual distribution of the liquid, which helps to more accurately estimate the volume of the liquid in subsequent steps and provide reliable data support for sterilization time compensation, etc.

[0057] Furthermore, the step S4 further includes:

[0058] S41: for the sterilized image after binarization, the number of pixels in the white area is counted, which is recorded as N;

[0059] S42: Calculate the liquid volume based on the binarized sterilization image. The calculation method is:

[0060] A=c·N;

[0061] V = W3·sigmoid(W2·A+b2);

[0062] Where A is the projected area of ​​the liquid on the image plane, c is the conversion coefficient from pixel to actual length, V is the liquid volume, W3 is the volume mapping matrix, W2 is the area projection mapping matrix, and b2 is the area projection bias vector;

[0063] S43: Calculate the liquid weight based on the liquid volume, the calculation method is:

[0064] M = V·ρ;

[0065] Where M is the weight of the liquid and ρ is the density coefficient of the liquid.

[0066] Furthermore, the step S5 further includes:

[0067] S51: According to the sterilizer manual, the initial sterilization time of the object to be sterilized is obtained, which is recorded as T;

[0068] S52: Combining the liquid weight and the liquid volume, a first compensation is performed on the initial sterilization time to obtain a first compensated sterilization time, which is calculated as follows:

[0069] T = MLP (M, V) + T;

[0070] Wherein, T' is the first compensation sterilization time;

[0071] S53: Combining the binarized sterilization image, performing a second compensation on the first compensated sterilization time to obtain a second compensated sterilization time, which is calculated as follows:

[0072] T=MLP(CNN(P binary ))+T;

[0073] Wherein, T” is the second compensation sterilization time;

[0074] S54: Starting a sterilization program according to the second compensated sterilization time to sterilize the object to be sterilized.

[0075] To further illustrate, the present invention performs feature extraction on the binarized sterilization image in step S53 to reflect the spatial distribution of the liquid, which serves as an important basis for determining the sterilization time compensation amount; if the second compensation is not performed in combination with the binarized sterilization image, then for liquids of equal quality but different spatial distribution, the algorithm may give the same sterilization time, resulting in insufficient sterilization time for large droplets and excessive sterilization time for multiple small droplets, affecting the sterilization effect and efficiency.

[0076] The present invention also discloses a medical sterilizer sterilization time compensation system, comprising:

[0077] Sterilization image acquisition and preprocessing module: acquires the original sterilization image and performs preprocessing to obtain the preprocessed sterilization image;

[0078] Sterilization image feature extraction module: Based on the pre-processed sterilization image, the liquid area segmentation image is calculated; then a morphological opening operation is performed to obtain a segmentation image after the opening operation; finally, feature extraction is performed on the pre-processed sterilization image to obtain the sterilization image features;

[0079] Liquid target detection module: liquid target recognition is performed based on the pre-processed sterilization image, the segmented image after the open operation and the sterilization image features to obtain a liquid target frame set; liquid target frames are then eliminated based on the sterilization image features to obtain a liquid target frame set after elimination; finally, the pre-processed sterilization image is binarized based on the liquid target frame set after elimination to obtain a binarized sterilization image;

[0080] Liquid volume and weight calculation module: for the binarized sterilized image, count the number of pixels in the white area and calculate the liquid volume and weight;

[0081] Sterilization time compensation module: obtains the initial sterilization time of the object to be sterilized, combines the liquid weight, liquid volume and the binarized sterilization image, performs sterilization time compensation and sterilizes the object to be sterilized.

[0082] Compared with the prior art, the present invention has the following beneficial effects:

[0083] (1) To address the problem of poor sterilization effect caused by insufficient consideration of liquid thermal inertia in traditional medical sterilizers, the present invention introduces the idea of ​​deep learning image processing, collects original sterilization images, performs feature extraction, calculates a set of liquid target frames, and removes inaccurate liquid target frames; then, the liquid volume and liquid weight are calculated based on the number of pixels in the liquid target frame, and the sterilization time is compensated in combination with the spatial distribution characteristics of the liquid, fully considering the influence of liquid thermal inertia, thereby improving the sterilization effect of the medical sterilizer.

[0084] (2) The present invention improves the U-Net network and designs an adaptive feature fusion module in the downsampling path. Through a multi-layer perceptron, the fusion coefficient is calculated by combining the feature map after convolution and the feature map after pooling, and the fusion ratio of the convolution feature and the pooling feature is dynamically adjusted, so that the network can better adapt to liquid areas of different scales and shapes; at the same time, the fusion of upsampling and downsampling features further improves the fineness and accuracy of the segmentation, and solves the problem that the traditional U-Net network is difficult to fully capture the association and difference between features in the sterilizer liquid target recognition scenario, thereby improving the accuracy of liquid target recognition and liquid volume and weight estimation.

[0085] (3) The present invention combines the liquid weight, liquid volume and the binarized sterilization image to perform sterilization time compensation. By comprehensively considering the properties and spatial distribution characteristics of the liquid, whether the liquid is in large flakes or distributed in multiple small flakes, the present invention can accurately identify and perform corresponding sterilization time compensation, thereby avoiding excess or insufficient sterilization time, improving sterilization efficiency, and solving the problem of poor sterilization effect caused by insufficient consideration of thermal inertia in traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 A schematic diagram of the process of the sterilization time compensation method for a medical sterilizer provided by the present invention;

[0087] Figure 2 A schematic diagram of the algorithm flow of the improved U-Net network provided by the present invention;

[0088] Figure 3 This is a schematic diagram of the algorithm flow of the sterilizer sterilization time compensation provided by the present invention. DETAILED DESCRIPTION

[0089] The present invention is further described below in conjunction with the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention belong to the protection scope of the present invention.

[0090] Embodiment 1: A method for compensating the sterilization time of a medical sterilizer, such as Figure 1 As shown, the following steps are included:

[0091] S1: Collect the original sterilization image and preprocess it to obtain the preprocessed sterilization image, including:

[0092] S11: Use a high-resolution camera at a fixed height to photograph the object to be sterilized from a top-down perspective to obtain an original sterilization image;

[0093] S12: preprocessing the original sterilization image, including: using an adaptive median filtering algorithm to perform image denoising, using a histogram equalization to perform image enhancement, and obtaining a sterilization image after preliminary preprocessing;

[0094] S13: geometrically correcting the sterilization image after preliminary preprocessing by using affine transformation to obtain a preprocessed sterilization image.

[0095] S2: Based on the pre-processed sterilization image, the liquid area segmentation image is calculated; then the morphological opening operation is performed to obtain the segmentation image after the opening operation; finally, the feature extraction of the pre-processed sterilization image is performed to obtain the sterilization image features, such as Figure 2 As shown, including:

[0096] S21: Based on the preprocessed sterilization image, an improved U-Net network is used for semantic segmentation to obtain a liquid area segmentation image, including:

[0097] Firstly, the preprocessed sterilization image is input into the improved U-Net network, which includes a downsampling path and an upsampling path. The upsampling path is used to obtain an upsampling feature map, and the downsampling path is used to obtain a downsampling fused feature map.

[0098] The improved U-Net network designs an adaptive feature fusion module in the downsampling path to adaptively fuse the feature map after convolution and the feature map after pooling. The calculation method is:

[0099]

[0100] in, is the feature map after downsampling and fusion of the Lth layer, L is the layer index, is the feature map after the Lth layer convolution, is the feature map after the Lth layer pooling, α (L) is the fusion coefficient of the Lth layer, sigmoid is the sigmoid function, MLP() is the multi-layer perceptron, and Concat() is the concatenation operation;

[0101] Then, the up-sampled feature map is concatenated with the down-sampled fused feature map, and the feature fusion is performed through a convolutional neural network to finally obtain the liquid area segmentation image. The calculation method is:

[0102]

[0103] Among them, P segmentation is the liquid area segmentation image, CNN() is the convolutional neural network, F up is the upsampled feature map;

[0104] S22: Using morphological opening operation, the liquid area segmentation image is first corroded and then expanded to obtain the segmentation image after opening operation, which is recorded as P' segmentation ;

[0105] S23: Extract features from the pre-processed sterilization image to obtain sterilization image features. The calculation method is:

[0106] F i =CNN(P,k i ,s i );

[0107] F feature =CNN 1×1 (Concat(F1,F2,…,F m));

[0108] att i =Attention(F i );

[0109]

[0110] Among them, F i is the sterilization image feature map of the i-th branch, i is the branch index, P is the preprocessed sterilization image, k i is the convolution kernel size of the i-th branch, s i is the convolution step size of the i-th branch, F feature For the preliminary sterilization image features, CNN 1×1 is a convolutional neural network with a convolution kernel size of 1×1, F1, F2, …, F m are the sterilization image feature maps of the 1st, 2nd, ..., mth branches respectively, m is the total number of branches, atti is the attention weight of the sterilization image feature map of the ith branch, Attention( ) is the attention mechanism, F' feature It is the sterilization image feature.

[0111] It should be noted that the U-Net network is a paper published by Ronneberger et al. in MICCAI, a top journal of computer medical imaging, in 2015. The paper first proposed a U-shaped structure for semantic segmentation of images.

[0112] The present invention improves the U-Net network and designs an adaptive feature fusion module in the downsampling path. The module calculates the fusion coefficient by combining the feature map after convolution and the feature map after pooling through a multi-layer perceptron. The adaptive feature fusion module can dynamically adjust the fusion ratio of convolution features and pooling features, so that the network can better adapt to liquid areas of different scales and shapes. At the same time, the fusion of upsampling and downsampling features further improves the precision and accuracy of segmentation.

[0113] In comparison, the traditional U-Net network only extracts features through convolution and pooling operations in the downsampling path, and restores the feature map size through deconvolution or upsampling operations in the upsampling path. Although this structure is simple, it is difficult to fully capture the association and difference between features when processing complex liquid areas, resulting in inaccurate segmentation results or missing important information, which in turn affects the accuracy of subsequent steps such as liquid target recognition, liquid volume and weight estimation.

[0114] In addition, the present invention adopts a multi-branch convolutional neural network structure, each branch uses a convolution kernel and step size of different sizes to extract features from the preprocessed sterilization image, which can capture features of different scales in the image and improve the diversity and richness of features; after feature extraction, a 1×1 convolutional neural network is used to fuse the feature maps of multiple branches. The 1×1 convolution kernel can adjust the number of channels of the feature map without changing the spatial size of the feature map, thereby realizing linear combination and dimensionality reduction of features; finally, an attention mechanism is introduced to weight the feature map of each branch, so that the model pays more attention to features that are beneficial to the task and suppresses irrelevant or noise features;

[0115] Compared with traditional neural networks, traditional methods usually use single-scale convolutional neural networks for feature extraction, which makes it difficult to capture features of different scales in the image. The sterilization image features extracted by the present invention are diverse and rich, and more accurately represent the liquid information in the image, which helps to more comprehensively analyze and process image data and improve the performance of the entire system.

[0116] S3: According to the pre-processed sterilization image, the segmented image after the opening operation and the sterilization image features, liquid target recognition is performed to obtain a liquid target frame set; then, the liquid target frame is eliminated according to the sterilization image features to obtain a liquid target frame set after elimination; finally, according to the liquid target frame set after elimination, the pre-processed sterilization image is binarized to obtain a binarized sterilization image, including:

[0117] S31: Liquid target recognition is performed based on the pre-processed sterilization image, the segmented image after the open operation, and the sterilization image features. The calculation method is:

[0118] B j =YOLOv8(ChannelConcat(P,P′ segmentation ,F′ feature ));

[0119] Among them, B j is the jth liquid target frame, j is the liquid target frame index, YOLOv8( ) is the target detection network, ChannelConcat( ) is the channel concatenation; B is calculated j Finally, all liquid target frames are uniformly stored to obtain a liquid target frame set;

[0120] S32: Eliminate the liquid target frame according to the sterilization image features to obtain a set of liquid target frames after elimination. The calculation process is as follows:

[0121] Firstly, in the preprocessed sterilization image, the image areas corresponding to all liquid target frames in the liquid target frame set are cropped, and the cropped areas are pixel-filled using the mean method to obtain the sterilization image background data;

[0122] Secondly, feature extraction is performed on the cropped area and sterilization image background data respectively, and the calculation method is as follows:

[0123]

[0124] f background =CNN(P background );

[0125] in, is the image region feature corresponding to the jth liquid target frame, LBP( ) is the local binary pattern, is the image area corresponding to the jth liquid target frame, GLCM( ) is the gray-level co-occurrence matrix, f background It is the background feature of sterilization image;

[0126] Finally, according to the background features of the sterilization image and the image area features corresponding to the liquid target frame, the liquid target frame rejection threshold and the target frame image area threshold are calculated respectively, and the liquid target frame is rejected. The calculation method is:

[0127] r=sigmoid(W1·f background +b1);

[0128]

[0129] Among them, r is the liquid target frame rejection threshold, W1 is the liquid target frame rejection threshold matrix, b1 is the liquid target frame rejection threshold offset vector, r j is the target frame image area threshold corresponding to the j-th liquid target frame;

[0130] Calculate r and r j After that, if r j >r, the liquid target frame is retained. If r j ≤r, the liquid target frame is removed;

[0131] After traversing and removing all liquid target frames, the remaining liquid target frames are uniformly stored to obtain a set of liquid target frames after removal;

[0132] S33: According to the eliminated liquid target frame set, the pre-processed sterilization image is binarized, the image area corresponding to the liquid target frame is filled with white, and the remaining area is filled with black, to obtain a binarized sterilization image, which is recorded as P binary .

[0133] It should be noted that the present invention performs feature extraction on the cropped liquid target frame area and the sterilization image background area respectively; uses LBP (local binary pattern) and GLCM (gray level co-occurrence matrix) to extract the texture features of the liquid target frame area, LBP generates a binary pattern by comparing the gray values ​​of a pixel with its neighboring pixels to capture local texture information; GLCM analyzes the gray value and spatial relationship of pixel pairs to quantify texture information; then, the liquid target frame rejection threshold and the target frame image area threshold are calculated respectively through linear transformation and sigmoid function;

[0134] Compared with traditional methods, most traditional methods only perform target frame removal based on pixel values ​​or simple morphological operations, which are easily affected by noise and complex background, resulting in false removal or missed removal; after removing inaccurate liquid target frames, the remaining liquid target frames more accurately represent the actual distribution of the liquid, which helps to more accurately estimate the volume of the liquid in subsequent steps and provide reliable data support for sterilization time compensation, etc.

[0135] For example, for the preprocessed sterilization image, after liquid target recognition in step S31, the following five liquid target frames are obtained, each of which is detected and located by the YOLOv8 network:

[0136] The boundary coordinates of liquid target frame 1 are: (53, 57, 102, 106);

[0137] The boundary coordinates of liquid target frame 2 are: (38, 42, 89, 94);

[0138] The boundary coordinates of liquid target frame 3 are: (72, 81, 120, 130);

[0139] The boundary coordinates of the liquid target frame 4 are: (15, 24, 60, 68);

[0140] The boundary coordinates of the liquid target frame 5 are: (91, 100, 145, 152);

[0141] The r value calculated by the liquid target frame rejection threshold is 0.7, and the target frame image area thresholds corresponding to each liquid target frame are: 0.85, 0.91, 0.65, 0.74, 0.63;

[0142] Since 0.65 and 0.63 are both smaller than 0.7, liquid target frame 3 and liquid target frame 5 should be eliminated, and liquid target frames 1, 2, and 4 should be retained as the liquid target frame set after elimination.

[0143] In particular, if the object to be sterilized is a glass product, before liquid target recognition in step S31, image enhancement and high-precision contour extraction are required to improve the accuracy of liquid recognition. The calculation method is:

[0144] P Contrast =ContrastStretching(P)

[0145] P Canny =CNN(Canny(ChannelConcat(P Contrast ,P)))

[0146] Among them, P Contrast is the image after contrast stretching, ContrastStretching( ) is contrast stretching, P Canny is the edge enhancement feature, and Canny() is the edge detection operator;

[0147] Calculate P Canny Afterwards, P Canny As part of the input to the YOLOv8 target detection network in step S31, the calculation formula of step S31 is updated as follows:

[0148] B j =YOLOv8(ChannelConcat(P,P′ segmentation ,F′ feature ,P Canny ));

[0149] The updated calculation formula is used to perform the calculation in step S31.

[0150] S4: Count the number of pixels in the white area of ​​the binarized sterilized image and calculate the liquid volume and liquid weight, including:

[0151] S41: for the sterilized image after binarization, the number of pixels in the white area is counted, which is recorded as N;

[0152] S42: Calculate the liquid volume based on the binarized sterilization image. The calculation method is:

[0153] A=c·N;

[0154] V = W3·sigmoid(W2·A+b2);

[0155] Where A is the projected area of ​​the liquid on the image plane, c is the conversion coefficient from pixel to actual length, V is the liquid volume, W3 is the volume mapping matrix, W2 is the area projection mapping matrix, and b2 is the area projection bias vector;

[0156] S43: Calculate the liquid weight based on the liquid volume, the calculation method is:

[0157] M = V·ρ;

[0158] Where M is the weight of the liquid and ρ is the density coefficient of the liquid.

[0159] S5: Obtain the initial sterilization time of the object to be sterilized, combine the liquid weight, liquid volume and the binarized sterilization image, perform sterilization time compensation and sterilize the object to be sterilized, such as Figure 3 As shown, including:

[0160] S51: According to the sterilizer manual, the initial sterilization time of the object to be sterilized is obtained, which is recorded as T;

[0161] S52: Combining the liquid weight and the liquid volume, a first compensation is performed on the initial sterilization time to obtain a first compensated sterilization time, which is calculated as follows:

[0162] T = MLP (M, V) + T;

[0163] Wherein, T' is the first compensation sterilization time;

[0164] S53: Combining the binarized sterilization image, performing a second compensation on the first compensated sterilization time to obtain a second compensated sterilization time, which is calculated as follows:

[0165] T″=MLP(CNN(P binary ))+T;

[0166] Wherein, T” is the second compensation sterilization time;

[0167] S54: Starting a sterilization program according to the second compensated sterilization time to sterilize the object to be sterilized.

[0168] To further illustrate, the present invention performs feature extraction on the binarized sterilization image in step S53 to reflect the spatial distribution of the liquid, which serves as an important basis for determining the sterilization time compensation amount; if the second compensation is not performed in combination with the binarized sterilization image, then for liquids of equal quality but different spatial distribution, the algorithm may give the same sterilization time, resulting in insufficient sterilization time for large droplets and excessive sterilization time for multiple small droplets, affecting the sterilization effect and efficiency.

[0169] For example, according to the standard operating procedures of high-temperature and high-pressure sterilization and the requirements of the sterilizer manual, the initial sterilization time of the scalpel is set to 20 minutes; there is a small amount of blood or saline remaining on the scalpel, with a total weight of 0.5 grams and a volume of 0.4 milliliters. After the first compensation, the compensation time output by the model is 3 minutes, and the first compensation sterilization time is 23 minutes; after the second compensation, due to the relatively concentrated spatial distribution of the residual liquid on the scalpel blade, the compensation time output by the model is 2 minutes, and the second compensation sterilization time is 25 minutes; finally, according to the second compensation sterilization time, the sterilization procedure is started, and the scalpel is sterilized at high temperature and high pressure to ensure that the scalpel is fully sterilized and meets the safety standards for medical use.

[0170] Embodiment 2: The present invention also discloses a sterilization time compensation system for a medical sterilizer, comprising:

[0171] Sterilization image acquisition and preprocessing module: acquires the original sterilization image and performs preprocessing to obtain the preprocessed sterilization image;

[0172] Sterilization image feature extraction module: Based on the pre-processed sterilization image, the liquid area segmentation image is calculated; then a morphological opening operation is performed to obtain a segmentation image after the opening operation; finally, feature extraction is performed on the pre-processed sterilization image to obtain the sterilization image features;

[0173] Liquid target detection module: liquid target recognition is performed based on the pre-processed sterilization image, the segmented image after the open operation and the sterilization image features to obtain a liquid target frame set; liquid target frames are then eliminated based on the sterilization image features to obtain a liquid target frame set after elimination; finally, the pre-processed sterilization image is binarized based on the liquid target frame set after elimination to obtain a binarized sterilization image;

[0174] Liquid volume and weight calculation module: for the binarized sterilized image, count the number of pixels in the white area and calculate the liquid volume and weight;

[0175] Sterilization time compensation module: obtains the initial sterilization time of the object to be sterilized, combines the liquid weight, liquid volume and the binarized sterilization image, performs sterilization time compensation and sterilizes the object to be sterilized.

[0176] The present invention is particularly suitable for sterilization time compensation calculation of surgical instruments (such as scalpels, surgical scissors, forceps, etc.), implants (such as artificial joints, pacemakers, etc.), catheters and cannulas (such as urinary catheters, endotracheal cannulas, etc.), and test tubes, beakers, culture dishes and other small items in medical laboratories; the present invention can effectively detect and process residual moisture or liquid on the surface of these small items through image processing methods, providing an accurate and reliable basis for subsequent sterilization time calculation, and is particularly suitable for sterilization preparation of the above-mentioned types of items.

[0177] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0178] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in multiple embodiments of the present invention.

[0179] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for compensating sterilization time of a medical sterilizer, characterized in that: The following steps are involved: S1: collecting the original sterilization image and preprocessing it to obtain the preprocessed sterilization image; S2: According to the pre-processed sterilization image, a liquid area segmentation image is calculated; then a morphological opening operation is performed to obtain a segmentation image after the opening operation; finally, feature extraction is performed on the pre-processed sterilization image to obtain sterilization image features; S3: According to the pre-processed sterilization image, the segmented image after the opening operation and the sterilization image features, liquid target recognition is performed to obtain a liquid target frame set; then, the liquid target frame is eliminated according to the sterilization image features to obtain a liquid target frame set after elimination; finally, according to the liquid target frame set after elimination, the pre-processed sterilization image is binarized to obtain a binarized sterilization image; S4: for the sterilized image after binarization, count the number of pixels in the white area, and calculate the liquid volume and liquid weight; S5: Obtain the initial sterilization time of the object to be sterilized, and perform sterilization time compensation and sterilization of the object to be sterilized based on the liquid weight, liquid volume and the binarized sterilization image.

2. The sterilization time compensation method based on a medical sterilizer according to claim 1, characterized in that: The S1 step includes: S11: Use a high-resolution camera at a fixed height to photograph the object to be sterilized from a top-down perspective to obtain an original sterilization image; S12: preprocessing the original sterilization image, including: using an adaptive median filtering algorithm to perform image denoising, using a histogram equalization to perform image enhancement, and obtaining a sterilization image after preliminary preprocessing; S13: geometrically correcting the sterilization image after preliminary preprocessing by using affine transformation to obtain a preprocessed sterilization image.

3. The sterilization time compensation method based on a medical sterilizer according to claim 1, characterized in that: The S2 step includes: S21: Based on the preprocessed sterilization image, an improved U-Net network is used for semantic segmentation to obtain a liquid area segmentation image, including: Firstly, the preprocessed sterilization image is input into the improved U-Net network, which includes a downsampling path and an upsampling path. The upsampling path is used to obtain an upsampling feature map, and the downsampling path is used to obtain a downsampling fused feature map. The improved U-Net network designs an adaptive feature fusion module in the downsampling path to adaptively fuse the feature map after convolution and the feature map after pooling. The calculation method is: in, is the feature map after downsampling and fusion of the Lth layer, L is the layer index, is the feature map after the Lth layer convolution, is the feature map after the Lth layer pooling, α (L) is the fusion coefficient of the Lth layer, sigmoid is the sigmoid function, MLP() is the multi-layer perceptron, and Concat() is the concatenation operation; Then, the up-sampled feature map is concatenated with the down-sampled fused feature map, and the feature fusion is performed through a convolutional neural network to finally obtain the liquid area segmentation image. The calculation method is: Among them, P segmentation is the liquid area segmentation image, CNN() is the convolutional neural network, F up is the upsampled feature map; S22: Using morphological opening operation, the liquid area segmentation image is first corroded and then expanded to obtain the segmentation image after opening operation, which is recorded as P' segmentation ; S23: Extract features from the pre-processed sterilization image to obtain sterilization image features. The calculation method is: F i =CNN(P,k i ,s i ); F feature =CNN 1×1 (Concat(F1,F2,…,F m )); to i =Attention(F i ); Among them, F i is the sterilization image feature map of the i-th branch, i is the branch index, P is the preprocessed sterilization image, k i is the convolution kernel size of the i-th branch, s i is the convolution step size of the i-th branch, F feature For the preliminary sterilization image features, CNN 1×1 is a convolutional neural network with a convolution kernel size of 1×1, F1, F2, …, F m are the sterilization image feature maps of the 1st, 2nd, ..., mth branches, respectively, m is the total number of branches, att i is the attention weight of the sterilization image feature map of the i-th branch, Attention() is the attention mechanism, F' feature It is the sterilization image feature.

4. The sterilization time compensation method based on a medical sterilizer according to claim 3, characterized in that: The S3 step includes: In the preprocessed sterilization image, the image areas corresponding to all liquid target frames in the liquid target frame set are cropped, and the cropped areas are pixel-filled using the mean method to obtain the sterilization image background data; The local binary pattern, gray-level co-occurrence matrix and multi-layer perceptron are combined to extract features of the image area corresponding to the liquid target frame; the convolutional neural network is used to extract features of the sterilization image background data to obtain the sterilization image background features; The sigmoid function is combined with the background features of the sterilization image to calculate the liquid target frame rejection threshold.

5. The sterilization time compensation method based on a medical sterilizer according to claim 3, characterized in that: The S3 step includes: S31: Liquid target recognition is performed based on the pre-processed sterilization image, the segmented image after the open operation, and the sterilization image features. The calculation method is: B j =YOLOv8(ChannelConcat(P,P′ segmentation ,F′ feature )); Among them, B j is the jth liquid target frame, j is the liquid target frame index, YOLOv8() is the target detection network, ChannelConcat() is the channel concatenation; B is calculated j Finally, all liquid target frames are uniformly stored to obtain a liquid target frame set; S32: Eliminate the liquid target frame according to the sterilization image features to obtain a set of liquid target frames after elimination. The calculation process is as follows: Firstly, in the preprocessed sterilization image, the image areas corresponding to all liquid target frames in the liquid target frame set are cropped, and the cropped areas are pixel-filled using the mean method to obtain the sterilization image background data; Secondly, feature extraction is performed on the cropped area and sterilization image background data respectively, and the calculation method is as follows: fbackground=CNN(Pbackground); in, is the image region feature corresponding to the jth liquid target frame, LBP() is the local binary pattern, is the image area corresponding to the jth liquid target frame, GLCM() is the gray-level co-occurrence matrix, and f background It is the background feature of sterilization image; Finally, according to the background features of the sterilization image and the image area features corresponding to the liquid target frame, the liquid target frame rejection threshold and the target frame image area threshold are calculated respectively, and the liquid target frame is rejected. The calculation method is: r=sigmoid(W1·f background +b1); Among them, r is the liquid target frame rejection threshold, W1 is the liquid target frame rejection threshold matrix, b1 is the liquid target frame rejection threshold offset vector, r j is the target frame image area threshold corresponding to the j-th liquid target frame; Calculate r and r j After that, if r j >r, the liquid target frame is retained. If r j ≤r, the liquid target frame is removed; After traversing and removing all liquid target frames, the remaining liquid target frames are uniformly stored to obtain a set of liquid target frames after removal; S33: According to the eliminated liquid target frame set, the pre-processed sterilization image is binarized, the image area corresponding to the liquid target frame is filled with white, and the remaining area is filled with black, to obtain a binarized sterilization image, which is recorded as P binary .

6. The sterilization time compensation method based on a medical sterilizer according to claim 4, characterized in that: The S4 step includes: S41: for the sterilized image after binarization, the number of pixels in the white area is counted, which is recorded as N; S42: Calculate the liquid volume based on the binarized sterilization image. The calculation method is: A=c·N; V = W3·sigmoid(W2·A+b2); Where A is the projected area of ​​the liquid on the image plane, c is the conversion coefficient from pixel to actual length, V is the liquid volume, W3 is the volume mapping matrix, W2 is the area projection mapping matrix, and b2 is the area projection bias vector; S43: Calculate the liquid weight based on the liquid volume, the calculation method is: M = V·ρ; Where M is the weight of the liquid and ρ is the density coefficient of the liquid.

7. The sterilization time compensation method based on a medical sterilizer according to claim 5, characterized in that: The S5 step includes: S51: According to the sterilizer manual, the initial sterilization time of the object to be sterilized is obtained, which is recorded as T; S52: Combining the liquid weight and the liquid volume, a first compensation is performed on the initial sterilization time to obtain a first compensated sterilization time, which is calculated as follows: T'=MLP(M,V)+T; Wherein, T' is the first compensation sterilization time; S53: Combining the binarized sterilization image, performing a second compensation on the first compensated sterilization time to obtain a second compensated sterilization time, which is calculated as follows: T″=MLP(CNN(P binary ))+T; Wherein, T” is the second compensation sterilization time; S54: Starting a sterilization program according to the second compensated sterilization time to sterilize the object to be sterilized.

8. A medical sterilizer sterilization time compensation system, characterized in that: include: Sterilization image acquisition and preprocessing module: acquires the original sterilization image and performs preprocessing to obtain the preprocessed sterilization image; Sterilization image feature extraction module: Based on the pre-processed sterilization image, the liquid area segmentation image is calculated; then a morphological opening operation is performed to obtain a segmentation image after the opening operation; finally, feature extraction is performed on the pre-processed sterilization image to obtain the sterilization image features; Liquid target detection module: liquid target recognition is performed based on the pre-processed sterilization image, the segmented image after the open operation and the sterilization image features to obtain a liquid target frame set; liquid target frames are then eliminated based on the sterilization image features to obtain a liquid target frame set after elimination; finally, the pre-processed sterilization image is binarized based on the liquid target frame set after elimination to obtain a binarized sterilization image; Liquid volume and weight calculation module: for the sterilized image after binarization, count the number of pixels in the white area and calculate the liquid volume and weight; Sterilization time compensation module: obtains the initial sterilization time of the object to be sterilized, combines the liquid weight, liquid volume and the binarized sterilization image, performs sterilization time compensation and sterilizes the object to be sterilized; To realize the sterilization time compensation method of a medical sterilizer as described in any one of claims 1-7.

Images