CN117392135A

CN117392135A - Injection dose analysis method and system based on image

Info

Publication number: CN117392135A
Application number: CN202311702727.XA
Authority: CN
Inventors: 张良; 凌兴鑫; 吴达; 文海
Original assignee: Shenzhen Pulang Medical Technology Development Co ltd
Current assignee: Shenzhen Pulang Medical Technology Development Co ltd
Priority date: 2023-12-12
Filing date: 2023-12-12
Publication date: 2024-01-12
Anticipated expiration: 2043-12-12
Also published as: CN117392135B

Abstract

The application relates to the technical field of image analysis and discloses an injection dose analysis method and system based on images. The method comprises the following steps: establishing an injection dose target tracking model; performing multi-layer feature recognition and multi-layer feature fusion to obtain a plurality of injection liquid level feature images; performing salient region identification to obtain a liquid level salient region, performing feature weight distribution calculation to obtain salient feature weights, and creating a first injection dose tracking strategy; carrying out regular weight calculation to obtain regular weight data and calculating association measurement and actual liquid level tracking position; performing consistency response and response difference calculation to obtain response difference data and performing strategy adjustment to obtain a second injection dose tracking strategy; and performing environmental change evaluation to obtain environmental change evaluation indexes and performing parameter dynamic adjustment to obtain a target injection dose tracking strategy of an injection dose target tracking model.

Description

Injection dose analysis method and system based on image

Technical Field

The present disclosure relates to the field of image analysis, and in particular, to an image-based injection dose analysis method and system.

Background

With the rapid development of artificial intelligence and image processing technology, medical image analysis using computer vision technology has become a trend. The injection dosage analysis method based on the images can realize accurate tracking of the injection liquid level by analyzing the image sequence of the injection equipment, so that not only can the errors of manual operation be reduced, but also the measurement speed and accuracy can be improved.

Conventional injection dose measurement methods rely mostly on manual observation and measurement, which is not only inefficient, but also susceptible to operator skill and experience, resulting in inaccuracy of the measurement results.

Disclosure of Invention

The application provides an injection dose analysis method and system based on images.

In a first aspect, the present application provides an image-based injection dose analysis method comprising:

collecting a target injection liquid level image sequence of injection equipment, extracting liquid level significant perception characteristics of the target injection liquid level image sequence to obtain liquid level significant perception characteristics, and establishing an injection dosage target tracking model according to the liquid level significant perception characteristics;

Performing multi-layer feature recognition and multi-layer feature fusion on the target injection liquid level image sequence to obtain a plurality of injection liquid level feature images;

performing salient region identification on the plurality of injection liquid level characteristic images respectively to obtain a liquid level salient region of each injection liquid level characteristic image, performing characteristic weight distribution calculation on the liquid level salient region to obtain salient characteristic weight of each injection liquid level characteristic image, and creating a first injection dose tracking strategy of the injection dose target tracking model according to the salient characteristic weight;

performing regular weight calculation on the injection liquid level characteristic images to obtain regular weight data, and calculating correlation metrics and actual liquid level tracking positions of the injection liquid level characteristic images according to the regular weight data;

performing consistent response and response difference calculation on the actual liquid level tracking position according to the injection dose target tracking model to obtain response difference data, and performing strategy adjustment on the first injection dose tracking strategy according to the response difference data to obtain a second injection dose tracking strategy;

and carrying out environmental change evaluation on the target injection liquid level image sequence to obtain an environmental change evaluation index, and carrying out parameter dynamic adjustment on the second injection dose tracking strategy according to the environmental change evaluation index to obtain a target injection dose tracking strategy of the injection dose target tracking model.

In a second aspect, the present application provides an image-based injection dose analysis system comprising:

the acquisition module is used for acquiring a target injection liquid level image sequence of the injection equipment, extracting liquid level significant perception characteristics of the target injection liquid level image sequence to obtain liquid level significant perception characteristics, and establishing an injection dosage target tracking model according to the liquid level significant perception characteristics;

the fusion module is used for carrying out multi-layer feature identification and multi-layer feature fusion on the target injection liquid level image sequence to obtain a plurality of injection liquid level feature images;

the identification module is used for respectively carrying out significance area identification on the plurality of injection liquid level characteristic images to obtain a liquid level significance area of each injection liquid level characteristic image, carrying out characteristic weight distribution calculation on the liquid level significance area to obtain a significant characteristic weight of each injection liquid level characteristic image, and creating a first injection dose tracking strategy of the injection dose target tracking model according to the significant characteristic weight;

the calculation module is used for carrying out regular weight calculation on the injection liquid level characteristic images to obtain regular weight data, and calculating the correlation measurement and the actual liquid level tracking position of the injection liquid level characteristic images according to the regular weight data;

The adjustment module is used for carrying out consistency response and response difference calculation on the actual liquid level tracking position according to the injection dose target tracking model to obtain response difference data, and carrying out strategy adjustment on the first injection dose tracking strategy according to the response difference data to obtain a second injection dose tracking strategy;

the evaluation module is used for carrying out environmental change evaluation on the target injection liquid level image sequence to obtain an environmental change evaluation index, and carrying out parameter dynamic adjustment on the second injection dose tracking strategy according to the environmental change evaluation index to obtain a target injection dose tracking strategy of the injection dose target tracking model.

In the technical scheme provided by the application, the accuracy of injection dose monitoring is effectively improved through obvious sensing feature extraction and multi-layer feature recognition. Compared with the traditional manual measurement, the method reduces human errors and ensures the accuracy and reliability of the measurement result. Through an automated image acquisition and analysis process, the method can track the change of the injection liquid level in real time and provide instant feedback. The adopted regular weight calculation and environment change evaluation mechanism enables the method to adapt to different environment conditions, such as illumination change, equipment movement and the like. This flexibility and adaptability allows for efficient operation in a variety of complex environments. Through consistency response calculation and response difference analysis, the change of the injection dosage can be tracked in real time, and the tracking strategy can be dynamically adjusted according to actual conditions. This real-time and flexibility allows the medical staff to respond in time to any changes in the injection process, ensuring the continuity and effectiveness of the treatment. By means of automatic analysis and tracking, the accuracy of injection dose analysis is improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained based on these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic representation of one embodiment of an image-based injection dose analysis method in an embodiment of the present application;

FIG. 2 is a schematic representation of one embodiment of an image-based injection dose analysis system in an embodiment of the present application.

Detailed Description

The embodiment of the application provides an image-based injection dose analysis method and system. The terms "first," "second," "third," "fourth" and the like in the description and in the claims of this application and in the above-described figures, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments described herein may be implemented in other sequences than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed or inherent to such process, method, article, or apparatus.

For ease of understanding, a specific flow of an embodiment of the present application is described below with reference to fig. 1, where an embodiment of an image-based injection dose analysis method of the embodiment of the present application includes:

step 101, acquiring a target injection liquid level image sequence of injection equipment, extracting liquid level significant perception features of the target injection liquid level image sequence to obtain liquid level significant perception features, and establishing an injection dosage target tracking model according to the liquid level significant perception features;

it will be appreciated that the subject of the present application may be an image-based injection dose analysis system, or may be a terminal or server, and is not limited in this regard. The embodiment of the present application will be described by taking a server as an execution body.

Specifically, first, an initial injection level image sequence of an injection device is acquired, and image data of the level is acquired using a high-precision image capturing device. Noise is contained in the acquired image sequence, so that the image quality is optimized through noise filtering processing, and the accuracy of subsequent analysis is ensured. Noise filtering involves the use of high pass, low pass or band pass filters to remove random noise or background interference from the image, resulting in a more clear and accurate sequence of target injection level images. Then, the saliency weight of the initial frame in the image sequence is calculated through a preset saliency weight function. A mathematical function is used to evaluate the saliency of each pixel in the image in order to highlight important features. The saliency weight function calculates the distance between each pixel point and the saliency center by adopting a Gaussian model, and saliency weight data is generated. The method can effectively identify the key area in the liquid level image and provide basis for subsequent feature extraction. And then, processing the image sequence weighted by the saliency weight data by adopting a feature extraction function to extract the liquid level saliency perception features. The liquid level significant perception feature is calculated by multiplying the significance weight of each pixel point by the feature vector thereof and then summing all the feature vectors. The weighted feature extraction method can emphasize key features in the image and improve the accuracy and efficiency of subsequent model analysis. And finally, establishing an initial target tracking model according to the extracted liquid level significant perception characteristics, and updating the model by utilizing a preset parameter updating function so as to obtain a target model for tracking injection dosage. The model typically includes a two-layer convolutional long short-time memory network (ConvLSTM), a pooling layer, and Sigmoid functions. ConvLSTM combines the ability of Convolutional Neural Networks (CNNs) to process spatial features with the advantage of long and short term memory networks (LSTMs) to process time series data, enabling models to effectively learn and track fluid level changes. The pooling layer helps to reduce the amount of computation and prevent overfitting, while the Sigmoid function is used to generate an output, such as a specific value or classification of the liquid level. With the structure, the model can continuously learn and adapt to new data, and tracking accuracy and robustness are improved.

102, carrying out multi-layer feature identification and multi-layer feature fusion on a target injection liquid level image sequence to obtain a plurality of injection liquid level feature images;

specifically, first, extracting a pre-set direction gradient Histogram (HOG) feature function to perform HOG feature extraction on each image frame in an image sequence, and calculating the gradient size and direction of each pixel point by using the HOG feature function, so as to form a gradient histogram of the image, thereby capturing and characterizing shape and texture information in the image. The HOG features are suitable for recognition and analysis of liquid surface textures in the liquid level image because they are sensitive to shape and texture changes of the partial image. Next, shallow feature extraction is performed on each image frame to obtain a shallow feature average value for each frame image. These shallow features include color histograms, edge information, or simple texture features that help describe the basic visual attributes of the image, which can help identify and distinguish liquid and non-liquid areas for a liquid level image. At the same time, mid-level feature extraction is also performed, which are more complex image attributes, such as Local Binary Pattern (LBP) or wavelet transform features, which help capture finer visual patterns in the image. Subsequently, by obtaining weighting parameters for the HOG features, shallow features, and middle features, these weighting parameters reflect the importance and contribution of the different features in the level analysis. On the basis, the HOG features, the shallow features and the middle features are subjected to weighted fusion calculation through a preset fusion feature function to form target fusion features. The feature fusion process is the core of the whole analysis method, because features of different layers and types are comprehensively considered, so that the accuracy and the robustness of liquid level identification are improved. The fused feature function effectively combines shape information of the HOG features, basic visual properties of the shallow features, and fine visual patterns of the middle features by giving each feature different weights. And finally, carrying out further feature processing on a plurality of image frames in the image sequence according to the obtained target fusion features to obtain a plurality of final injection liquid level feature images. The characteristic images integrate the characteristics extracted from different layers, describe the state of the liquid level more comprehensively and accurately, and provide a reliable basis for subsequent dose analysis and tracking. Through the comprehensive analysis and fusion of the multi-level characteristics, the change of the liquid level can be more effectively identified and tracked, so that the accurate metering of the injection dosage is ensured.

Step 103, performing salient region identification on a plurality of injection liquid level characteristic images respectively to obtain a liquid level salient region of each injection liquid level characteristic image, performing characteristic weight distribution calculation on the liquid level salient region to obtain a salient characteristic weight of each injection liquid level characteristic image, and creating a first injection dose tracking strategy of an injection dose target tracking model according to the salient characteristic weights;

specifically, first, the saliency area identification is performed on the plurality of injection liquid level characteristic images respectively to determine the saliency area of the liquid level in each image. Those areas of the entire image sequence that are most critical for fluid level monitoring are identified by image processing algorithms. This involves the use of edge detection, texture analysis, color segmentation, etc. techniques to accurately locate the boundaries and surfaces of the liquid. Next, weight calculation is performed on these salient regions by a preset salient perceptual weight function to obtain salient perceptual weights. The weight function adopts an exponential decay model based on distance, and the weight is distributed by calculating the distance from the feature point to the obvious center, so that the feature point close to the obvious center obtains higher weight. This weighting method helps to highlight the most important features in the level monitoring while suppressing those areas that are not much related to the level change, thereby improving the accuracy and efficiency of the overall analysis. Further, the liquid level significance region is subjected to feature adjustment according to the significance perception weight, and adjusted features are obtained. The importance of the salient region is reflected by weighting the original features by multiplying them by the salient perceptual weights. By means of the adjustment, the model can be ensured to pay more attention to the features which are most critical to liquid level monitoring when analyzing and processing image data, and therefore the accuracy and response speed of the model are improved. And then, carrying out weight normalization processing on the adjusted characteristics through a preset weight normalization function to obtain the significant characteristic weight of each injection liquid level characteristic image. Weight normalization ensures that the comparison and analysis-by-synthesis between different features is performed under a unified standard. The normalization process helps to eliminate scale differences between different features so that the model can more equitably evaluate the effect of each feature on the level change. And finally, setting an influence coefficient of the adjusted feature in the injection dose target tracking model according to the significant feature weight, and generating a corresponding first injection dose tracking strategy according to the influence coefficient. The significant feature weights are converted into influence coefficients in the model, thereby guiding the model how to weigh the importance of each feature when processing different image frames.

104, carrying out regular weight calculation on the plurality of injection liquid level characteristic images to obtain regular weight data, and calculating correlation metrics and actual liquid level tracking positions of the plurality of injection liquid level characteristic images according to the regular weight data;

specifically, first, the target liquid level positions of two adjacent image frames are compared based on a preset regular weight function, and regular weight data is calculated. The consistency of the liquid level positions in adjacent image frames is evaluated, and the regular weight function gives different weights by considering the degree of change of the positions. The method can effectively identify the degree of liquid level position change and is beneficial to follow-up liquid level tracking and analysis. Then, according to the obtained regular weight data, the correlation metrics of the injection liquid level characteristic images are further calculated. The purpose of the correlation metric function is to evaluate the similarity or variability of features between successive image frames, calculated by comparing the feature vector differences of adjacent frames, and combining the canonical weight data. Such a correlation metric takes into account not only the variability of the features, but also the consistency of these differences over time, which helps to accurately track the level change. Finally, an actual level tracking position is determined based on the correlation metric by a preset position calculation function. The position calculation function not only considers the liquid level position between the current frame and the previous frame, but also incorporates a correlation metric to adjust the liquid level tracking position. In this process, the adjustment factor serves to balance the level variation between successive frames, ensuring accuracy and consistency of level tracking. In this way, variations in injection level can be tracked more accurately, providing accurate injection dose monitoring in medical applications.

Step 105, carrying out consistent response and response difference calculation on the actual liquid level tracking position according to the injection dose target tracking model to obtain response difference data, and carrying out strategy adjustment on the first injection dose tracking strategy according to the response difference data to obtain a second injection dose tracking strategy;

specifically, firstly, according to an injection dose target tracking model, liquid level tracking position prediction is carried out on a plurality of injection liquid level characteristic images to obtain a target liquid level tracking position, liquid level characteristics are analyzed by utilizing image processing and a machine learning algorithm, and the change trend of liquid level is predicted. The accuracy of the target tracking model directly affects the reliability of the liquid level prediction. And then, based on a preset consistency response function, carrying out consistency response calculation on the actual liquid level tracking position and the target liquid level tracking position predicted by the model. The purpose is to evaluate the consistency between the actual liquid level position and the predicted position of the model, so as to judge the accuracy and the reliability of the model. The consistency response function generates a consistency response indicator by calculating the difference between the actual position and the predicted position, which helps to quantify the accuracy of the model predictions. And then, acquiring an actual consistency response index of the injection dose target tracking model, and calculating response difference data between the actual consistency response index and the target consistency response index through a preset response difference function. This can reveal the deviation of the model from the ideal state in practical applications. The response difference data provides important information about the performance of the model, helping to identify problems with the model during the level tracking process. Finally, the first injection dose tracking strategy is adjusted according to the calculated response difference data, so that an improved second injection dose tracking strategy is obtained. The purpose of strategy adjustment is to optimize the accuracy and response capability of the model, ensuring a more accurate and reliable liquid level tracking process. This involves adjusting parameters of the model, such as learning rate, feature selection, algorithm structure, etc., or introducing new data processing and analysis techniques to improve the overall performance of the model. Through the continuous optimization and adjustment, the injection dosage tracking strategy can be more in line with the requirements of practical application, and the accurate monitoring of the liquid level in the injection process is ensured.

And 106, performing environmental change evaluation on the target injection liquid level image sequence to obtain an environmental change evaluation index, and performing parameter dynamic adjustment on the second injection dose tracking strategy according to the environmental change evaluation index to obtain a target injection dose tracking strategy of the injection dose target tracking model.

Specifically, first, a target injection level image sequence is analyzed by a preset environmental change evaluation function to evaluate the influence of environmental factors such as illumination change, equipment movement and the like on image quality and level identification. The environmental change evaluation function determines the degree of change in the environment mainly by comparing the difference in feature vectors of successive image frames. By calculating the difference of the feature vectors of the front frame and the rear frame and giving proper weight factors, the influence of the environmental change on the liquid level identification can be effectively quantified, and an environmental change evaluation index can be generated. And then, utilizing a preset parameter dynamic adjustment function to carry out parameter adjustment on the second injection dosage tracking strategy according to the environmental change evaluation index. And dynamically adjusting parameters of the tracking model according to the degree and the direction of the environmental change so as to adapt to the influence caused by the environmental change. The parameter dynamic adjustment function adjusts model parameters by considering the environmental change evaluation index and the difference between the current liquid level position and the previous position, and generates new target dynamic tracking parameters. The dynamic adjustment mechanism enables the tracking strategy to flexibly adapt to environmental changes, so that the accuracy and stability of liquid level tracking are improved. And finally, further dynamically adjusting the second injection dose tracking strategy according to the obtained target dynamic tracking parameters to form a final injection dose target tracking model. Ensuring that the tracking strategy is not only effective in the current environment, but also has the ability to adapt to future environmental changes. By the mode, the injection dosage target tracking model can accurately monitor and track the change of the liquid level, and the accuracy of dosage in the injection process is ensured. In addition, the dynamic adjustment strategy can also reduce interference of environmental factors on liquid level identification, and improve the robustness and reliability of the whole system.

In the embodiment of the application, the accuracy of injection dose monitoring is effectively improved through the obvious sensing feature extraction and the multi-layer feature recognition. Compared with the traditional manual measurement, the method reduces human errors and ensures the accuracy and reliability of the measurement result. Through an automated image acquisition and analysis process, the method can track the change of the injection liquid level in real time and provide instant feedback. The adopted regular weight calculation and environment change evaluation mechanism enables the method to adapt to different environment conditions, such as illumination change, equipment movement and the like. This flexibility and adaptability allows for efficient operation in a variety of complex environments. Through consistency response calculation and response difference analysis, the change of the injection dosage can be tracked in real time, and the tracking strategy can be dynamically adjusted according to actual conditions. This real-time and flexibility allows the medical staff to respond in time to any changes in the injection process, ensuring the continuity and effectiveness of the treatment. By means of automatic analysis and tracking, the accuracy of injection dose analysis is improved.

In a specific embodiment, the process of executing step 101 may specifically include the following steps:

(1) Acquiring an initial injection liquid level image sequence of injection equipment, and performing noise filtering treatment on the initial injection liquid level image sequence to obtain a target injection liquid level image sequence;

(2) Performing saliency weight calculation on an initial frame in a target injection liquid level image sequence through a preset saliency weight function to obtain saliency weight data, wherein the saliency weight function is as follows:，/>representing significance weight data, +.>Is the current pixel point, +.>Is a significant center, is->Is the standard deviation of the control significance area range, +.>Is a normalization factor;

(3) Performing saliency weighted feature extraction on an initial frame in a target injection liquid level image sequence according to saliency weight data through a preset feature extraction function to obtain liquid level saliency perception features, wherein the feature extraction function is as follows: ，/>representing a significant perceived characteristic of the fluid level->Is a feature vector, ++>Is the number of features;

(4) Establishing an initial target tracking model according to the liquid level significant perception characteristics, and updating model parameters of the initial target tracking model according to a preset parameter updating function to obtain an injection dose target tracking model, wherein the parameter updating function is as follows:,/>an injection dose target tracking model representing the current moment, < + > >An initial object tracking model representing the previous moment, < +.>Is a smoothing parameter +.>The liquid level significant perception feature is represented, and the injection dose target tracking model comprises a two-layer convolution long short-time memory network, a pooling layer and a Sigmiod function.

Specifically, first, an initial injection level image sequence of an injection device is acquired, and a change in injection level is captured using a high-precision imaging device. These images are then subjected to noise filtering to improve image quality and reduce errors in subsequent analysis. Noise filtering typically involves the use of digital image processing techniques, such as gaussian or median filters, to eliminate random noise in the image and preserve critical image features. Then, the initial frame in the target injection liquid level image sequence is subjected to saliency weight calculation through a preset saliency weight function, and the most critical area for liquid level monitoring in the image is identified. The saliency weight function highlights the most important parts of the image for liquid level monitoring by calculating the distance between each pixel point and the saliency center and using a gaussian model to give different weights to these distances. The method can effectively concentrate the analysis key point on the area with the most obvious liquid level change, and lays a foundation for the subsequent liquid level analysis. And then, performing saliency weighted feature extraction on the initial frames in the image sequence according to the saliency weight data through a preset feature extraction function. And analyzing the weighted image by using an algorithm, and extracting the obvious perception characteristics of the liquid level. The feature extraction function typically combines the saliency weights with feature vectors in the image to generate a liquid level saliency-aware feature by computing a weighted sum of the different features. Such feature extraction methods can ensure that the model focuses on those most critical image areas when analyzing level changes. And then, establishing an initial target tracking model according to the extracted liquid level significant perception characteristics, and carrying out parameter updating on the model by utilizing a preset parameter updating function so as to obtain an injection dose target tracking model. Machine learning or deep learning techniques are used to learn and analyze the level salient perceptual features. The parameter updating function of the model can continuously adjust the parameters of the model according to the newly extracted characteristics, so that the model can adapt to the new characteristics of liquid level change. Parameter updates typically employ a smooth transition, such as a weighted average, to enable the model to transition smoothly from an initial state to a more accurate tracking state.

In a specific embodiment, the process of executing step 102 may specifically include the following steps:

(1) Performing HOG feature extraction on a plurality of image frames in a target injection liquid level image sequence through a preset HOG feature function to obtain HOG features of each image frame, wherein the HOG feature function is as follows:h represents HOG characteristics, < >>Is the gradient in the x-direction, +.>Is the gradient in the y direction;

(2) Shallow feature extraction is carried out on a plurality of image frames in a target injection liquid level image sequence respectively to obtain a shallow feature average value of each image frame, wherein the shallow features are as follows:，/>is a single shallow feature, +.>The average value of the shallow features is N, and the number of the shallow features is N;

(3) Respectively extracting middle layer characteristics of a plurality of image frames in a target injection liquid level image sequence to obtain middle layer characteristics of each image frame;

(4) Respectively acquiring a first weight parameter of the HOG feature, a second weight parameter of the average value of the shallow feature and a third weight parameter of the middle layer feature;

(5) According to the first weight parameter, the second weight parameter and the third weight parameter, carrying out feature fusion calculation on the HOG feature, the shallow feature average value and the middle layer feature through a preset fusion feature function to obtain a target fusion feature, wherein the fusion feature function is as follows: ，/>Representing the target fusion feature->Represents a first weight parameter, H represents HOG characteristics,>representing a second weight parameter,/->Is the average value of shallow features, +.>Representing a third weight parameter, ++>Representing middle layer characteristics;

(6) And carrying out feature processing on a plurality of image frames in the target injection liquid level image sequence according to the target fusion features to obtain a plurality of injection liquid level feature images.

Specifically, first, a pre-set Histogram of Oriented Gradient (HOG) feature function is used to perform HOG feature extraction on each image frame in the target injection level image sequence. HOG feature extraction is performed by calculating the gradient of each pixel in the image in the x and y directions, and then generating a gradient histogram of the image based on the gradient information. Edge detection algorithms, such as Sobel or Canny operators, are used to identify edges and texture information in the image. By calculating the gradient magnitude and direction of each pixel, HOG features can effectively capture shape and structure information of an image. And then, shallow feature extraction is carried out on each image frame so as to obtain an average value of the shallow features. Shallow features typically include basic visual properties of an image such as color histogram, brightness, and contrast. By calculating the average of these shallow features, a representation of the global visual properties of the summarized image can be obtained. These shallow features help identify the overall visual pattern aspect of the image. The image frames are then subjected to mid-level feature extraction to obtain more complex image attributes. The mid-layer features include Local Binary Patterns (LBP), wavelet transform features, or other texture features. These mid-level features are typically capable of capturing advanced visual patterns in the image, such as texture and shape changes, to accurately identify and track liquid level changes. And then, acquiring weight parameters of the HOG features, the shallow features and the middle features, and carrying out weighted fusion on the features of different layers through a preset fusion feature function. The purpose of the fusion feature function is to comprehensively consider the importance and contribution of different features in liquid level monitoring and generate a comprehensive target fusion feature. The feature fusion method is beneficial to centralizing the advantages of various features and reducing the limitation caused by single features. And finally, carrying out further feature processing on a plurality of image frames in the image sequence according to the obtained target fusion feature so as to obtain a final injection liquid level feature image. The characteristic images integrate the characteristics extracted from different layers, can describe the state of the liquid level more comprehensively and accurately, and provide a reliable basis for subsequent dose analysis and tracking.

In a specific embodiment, the process of executing step 103 may specifically include the following steps:

(1) Performing salient region identification on the injection liquid level characteristic images respectively to obtain a liquid level salient region of each injection liquid level characteristic image;

(2) Performing significant perception weight calculation on the liquid level significant region through a preset significant perception weight function to obtain significant perception weight, wherein the significant perception weight function is：,/>Representing significant perceptual weights, ++>Representing the distance of the feature point to the salient center;

(3) According to the significant perception weight, carrying out characteristic adjustment on the liquid level significant region to obtain an adjusted characteristic, wherein the adjusted characteristic calculation formula is as follows:,/>representing the adjusted characteristics, ++>Representing significant perceptual weights, ++>Representing the original features;

(4) Carrying out weight normalization on the adjusted features through a preset weight normalization function to obtain the significant feature weight of each injection liquid level feature image, wherein the weight normalization function is as follows:,/>representing salient feature weights, ++>Representing the adjusted feature;

(5) And setting an influence coefficient of the adjusted characteristic in the injection dose target tracking model according to the significant characteristic weight, and generating a corresponding first injection dose tracking strategy according to the influence coefficient.

Specifically, firstly, saliency region identification is performed on a plurality of injection liquid level characteristic images respectively, and a region which is most critical to liquid level change in the images is determined. Significant areas of the fluid level are identified and marked using advanced image processing techniques such as edge detection, image segmentation, or deep learning models. The identification of the salient region determines the important region of analysis, and ensures that the subsequent processing is concentrated on the most significant part of the liquid level change. Next, the identified salient regions are subjected to salient perceptual weighting calculations by a preset salient perceptual weighting function. The distance from each feature point to the salient center is calculated and weights are assigned based on this distance. The significant perceptual weighting function typically employs a distance-based decay model, such as a gaussian function, to determine the weight of each feature point. This distance-based weight distribution ensures that feature points near the center of significance get higher weights, highlighting the areas where the level change is most significant. And then, according to the calculated significant perceived weight, carrying out characteristic adjustment on the liquid level significant region. The contribution of salient regions in the feature representation is enhanced by combining the original features with salient perceptual weights. The adjusted characteristics can more accurately reflect the key information of the liquid level change. The feature adjustment process typically involves multiplying the original feature value for each feature point by a corresponding significant perceptual weight, thereby generating an adjusted feature representation. And then, carrying out weight normalization on the adjusted characteristics through a preset weight normalization function to obtain the significant characteristic weight of each injection liquid level characteristic image. The purpose of the weight normalization is to ensure that the characteristic representations of the different image frames are compared and analyzed under the same standard, thereby eliminating differences in the scale of the characteristic values. The normalized salient feature weights provide a uniform, comparable feature standard for subsequent analysis. And finally, setting an influence coefficient of the adjusted feature in the injection dose target tracking model according to the obtained significant feature weight, and generating a corresponding first injection dose tracking strategy according to the influence coefficient. The results of the feature analysis are converted into specific tracking strategies, and feature weights are converted into influence coefficients in the model so as to guide the model in how to weigh the importance of each feature when processing different image frames.

In a specific embodiment, the process of executing step 104 may specifically include the following steps:

(1) Based on a preset regular weight function, carrying out regular weight calculation on two adjacent image frames in the multiple injection liquid level characteristic images to obtain regular weight data,，/>representing canonical weight data reflecting the consistency of target level positions of two adjacent image frames, +.>Adjustment coefficient representing position difference, +.>Target level position representing the video frame of the previous image, is->Representing a target liquid level position of a current image video frame;

(2) Calculating a correlation metric of the plurality of injection level feature images according to the canonical weight data, wherein the correlation metric function is:，/>feature vector representing the video frame of the previous image, is->Feature vector representing the video frame of the current image, +.>Representing the association metric +.>Representing canonical weight data, +.>An adjustment coefficient representing the characteristic difference;

(3) Calculating actual liquid level tracking positions of a plurality of injection liquid level characteristic images according to the correlation measurement by a preset position calculation function, wherein the position calculation function is as follows:，/>target level position representing the video frame of the previous image, is->Target level position representing the current image video frame, is- >Is a regulator, and is a->Representing a correlation measure between two adjacent frames, +.>Representing the actual level tracking position.

Specifically, firstly, performing regular weight calculation on two adjacent image frames in a plurality of injection liquid level characteristic images through a preset regular weight function, and evaluating the consistency of liquid level positions in the adjacent image frames. The canonical weight function assigns weights by taking into account the degree of difference in target level locations between adjacent two image frames. The weight distribution method based on the position difference can effectively identify and emphasize consistency or inconsistency of liquid level position change, and provides important reference basis for subsequent liquid level tracking. Next, based on the obtained canonical weight data, a correlation metric for the plurality of injection level feature images is calculated. The purpose of the correlation metric function is to measure the similarity or difference of features between successive image frames. By comparing the feature vector differences of adjacent frames and combining regular weight data to perform calculation, consistency of liquid level change on time sequence can be effectively evaluated. The computation of the correlation metric not only takes into account the variability of the feature vectors, but also combines the consistency of these differences over time, thereby providing critical information for accurately tracking the level changes. Finally, determining the actual liquid level tracking position according to the correlation measurement through a preset position calculation function. The previous analysis results are converted into specific liquid level position information, wherein the position calculation function combines the liquid level position difference between the current frame and the previous frame and the correlation measure to adjust the liquid level tracking position. The method can consider the dynamic property of liquid level change and ensure the accuracy and the continuity of liquid level tracking.

In a specific embodiment, the process of executing step 105 may specifically include the following steps:

(1) Performing liquid level tracking position prediction on a plurality of injection liquid level characteristic images according to an injection dose target tracking model to obtain a target liquid level tracking position;

(2) Based on a preset consistency response function, carrying out consistency response calculation on the actual liquid level tracking position and the target liquid level tracking position to obtain a target consistency response index, wherein the consistency response function is as follows:，is the actual level tracking position, +.>Is the target level tracking position,/->Factor representing the steepness of the control response curve +.>Representing a target consistency response index;

(3) Acquiring an actual consistency response index of an injection dose target tracking model, and calculating an actual consistency response index through a preset response difference functionAnd the response difference data of the target and the target consistency response index, wherein the response difference function is as follows:，/>representing response difference data, ++>Indicating that the target is consistent with the response index,representing an actual consistency response index;

(4) And carrying out strategy adjustment on the first injection dose tracking strategy according to the response difference data to obtain a second injection dose tracking strategy.

Specifically, first, liquid level tracking position prediction is performed on a plurality of injection liquid level characteristic images according to an injection dose target tracking model. By using image processing and machine learning algorithms, such as Convolutional Neural Networks (CNNs) or long-term memory networks (LSTM), the liquid level characteristics are analyzed and the trend of the liquid level change is predicted. Through deep analysis of historical and real-time image data, the model can accurately predict the liquid level position at a certain moment in the future, and provides a basis for accurate control of injection dosage. And then, based on a preset consistency response function, carrying out consistency response calculation on the actual liquid level tracking position and the target liquid level tracking position predicted by the model. The consistency response function is designed to evaluate consistency between the actual level position and the model predicted position. By calculating the difference between the two and applying a factor that controls the sharpness of the response curve, a target consistency response index can be generated. The index reflects the coincidence degree between the actual liquid level position and the predicted position and is a key index for evaluating the prediction accuracy of the model. And then, acquiring an actual consistency response index of the injection dose target tracking model, and calculating response difference data between the actual consistency response index and the target consistency response index through a preset response difference function. The effect of the response difference function is to quantify the deviation between the actual response and the ideal response, thereby providing important information about the performance of the model. These response difference data are the basis for evaluating and adjusting the injected dose tracking strategy, and they reveal the gap between the model's behavior in practical applications and the ideal state. Finally, the first injection dose tracking strategy is adjusted according to the calculated response difference data to form an improved second injection dose tracking strategy. And adjusting and optimizing parameters and logic of the model according to the performance of the model in practical application so as to improve the accuracy and reliability of the model. Through the continuous optimization and adjustment, the injection dosage tracking strategy can be more in line with the requirements of practical application, and the accurate monitoring of the liquid level in the injection process is ensured.

In a specific embodiment, the process of executing step 106 may specifically include the following steps:

(1) Performing environmental change evaluation on the target injection liquid level image sequence through a preset environmental change evaluation function to obtain an environmental change evaluation index, wherein the environmental change evaluation function is as follows:，/>indicating an environmental change assessment indicator->Weight factor representing the change of the evaluation environment, +.>Feature vector representing the video frame of the previous image, is->A feature vector representing a video frame of the current image;

(2) And carrying out parameter dynamic adjustment calculation on the second injection dose tracking strategy according to the environmental change evaluation index by a preset parameter dynamic adjustment function to obtain a target dynamic tracking parameter, wherein the parameter dynamic adjustment function is as follows:，/>indicating an environmental change assessment indicator->Target level position representing the video frame of the previous image, is->Target level position representing the current image video frame, is->Coefficients representing the adjustment of the tracking parameters +.>Representing a dynamic tracking parameter of the target; (3) And carrying out parameter dynamic adjustment on the second injection dose tracking strategy according to the target dynamic tracking parameters to obtain a target injection dose tracking strategy of the injection dose target tracking model.

Specifically, first, an environmental change evaluation is performed on a target injection liquid level image sequence through a preset environmental change evaluation function, and environmental factors such as illumination change, equipment vibration or temperature change affecting liquid level monitoring are identified and quantified. The environmental change evaluation function determines the change condition of the environment by calculating the degree of change of the feature vector between the successive image frames. By giving a weight factor to the feature vector change of each frame, the influence degree of the environmental change on the liquid level monitoring can be quantified, and an environmental change evaluation index can be generated. And then, carrying out parameter dynamic adjustment calculation on the second injection dose tracking strategy according to the obtained environmental change evaluation index by a preset parameter dynamic adjustment function. Parameters of the injection dose target tracking model are adjusted to adapt to the influence caused by environmental changes. The parameter dynamic adjustment function adjusts model parameters by considering the environmental change evaluation index and the difference between the current liquid level position and the previous position, and generates new target dynamic tracking parameters. The dynamic adjustment mechanism enables the tracking strategy to flexibly adapt to environmental changes, so that the accuracy and stability of liquid level tracking are improved. And finally, further dynamically adjusting the second injection dose tracking strategy according to the obtained target dynamic tracking parameters to form a final injection dose target tracking model. Ensuring that the tracking strategy is not only effective in the current environment, but also has the ability to adapt to future environmental changes. By the mode, the injection dosage target tracking model can accurately monitor and track the change of the liquid level, and the accuracy of dosage in the injection process is ensured. In addition, the dynamic adjustment strategy can also reduce interference of environmental factors on liquid level identification, and improve the robustness and reliability of the whole system.

The method for analyzing an injection dose based on an image in the embodiment of the present application is described above, and the system for analyzing an injection dose based on an image in the embodiment of the present application is described below, referring to fig. 2, an embodiment of the system for analyzing an injection dose based on an image in the embodiment of the present application includes:

the acquisition module 201 is configured to acquire a target injection liquid level image sequence of an injection device, extract liquid level significant perceptual features of the target injection liquid level image sequence, obtain liquid level significant perceptual features, and establish an injection dose target tracking model according to the liquid level significant perceptual features;

the fusion module 202 is configured to perform multi-layer feature recognition and multi-layer feature fusion on the target injection liquid level image sequence to obtain a plurality of injection liquid level feature images;

the identifying module 203 is configured to identify the salient regions of the plurality of injection liquid level feature images, obtain a liquid level salient region of each injection liquid level feature image, perform feature weight distribution calculation on the liquid level salient region, obtain a salient feature weight of each injection liquid level feature image, and create a first injection dose tracking strategy of the injection dose target tracking model according to the salient feature weights;

The calculation module 204 is configured to perform regular weight calculation on the plurality of injection liquid level feature images to obtain regular weight data, and calculate correlation metrics and actual liquid level tracking positions of the plurality of injection liquid level feature images according to the regular weight data;

the adjustment module 205 is configured to perform consistent response and response difference calculation on the actual liquid level tracking position according to the injection dose target tracking model to obtain response difference data, and perform policy adjustment on the first injection dose tracking policy according to the response difference data to obtain a second injection dose tracking policy;

the evaluation module 206 is configured to perform environmental change evaluation on the target injection level image sequence to obtain an environmental change evaluation index, and perform parameter dynamic adjustment on the second injection dose tracking strategy according to the environmental change evaluation index to obtain a target injection dose tracking strategy of the injection dose target tracking model.

Through the cooperation of the components, the accuracy of injection dose monitoring is effectively improved through obvious sensing feature extraction and multi-layer feature recognition. Compared with the traditional manual measurement, the method reduces human errors and ensures the accuracy and reliability of the measurement result. Through an automated image acquisition and analysis process, the method can track the change of the injection liquid level in real time and provide instant feedback. The adopted regular weight calculation and environment change evaluation mechanism enables the method to adapt to different environment conditions, such as illumination change, equipment movement and the like. This flexibility and adaptability allows for efficient operation in a variety of complex environments. Through consistency response calculation and response difference analysis, the change of the injection dosage can be tracked in real time, and the tracking strategy can be dynamically adjusted according to actual conditions. This real-time and flexibility allows the medical staff to respond in time to any changes in the injection process, ensuring the continuity and effectiveness of the treatment. By means of automatic analysis and tracking, the accuracy of injection dose analysis is improved.

It will be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, systems and units may refer to the corresponding processes in the foregoing method embodiments, which are not repeated herein.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be embodied in essence or a part contributing to the prior art or all or part of the technical solution in the form of a software product stored in a storage medium, including several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a read-only memory (ROM), a random access memory (random access memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

The above embodiments are merely for illustrating the technical solution of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the corresponding technical solutions.

Claims

1. An image-based injection dose analysis method, characterized in that the image-based injection dose analysis method comprises:

2. The method for analyzing injection dose based on image according to claim 1, wherein the steps of acquiring a target injection liquid level image sequence of an injection device, extracting liquid level significant perceptual features of the target injection liquid level image sequence, obtaining liquid level significant perceptual features, and establishing an injection dose target tracking model according to the liquid level significant perceptual features comprise:

Acquiring an initial injection liquid level image sequence of injection equipment, and performing noise filtering treatment on the initial injection liquid level image sequence to obtain a target injection liquid level image sequence;

performing saliency weight calculation on an initial frame in the target injection liquid level image sequence through a preset saliency weight function to obtain saliency weight data, wherein the saliency weight function is as follows:，/>representing significance weight data, +.>Is the current pixel point, +.>Is a significant center, is->Is the standard deviation of the control significance area range, +.>Is a normalization factor;

performing saliency weighted feature extraction on an initial frame in the target injection liquid level image sequence according to the saliency weight data through a preset feature extraction function to obtain a liquid level saliency perception feature, wherein the feature extraction function is as follows:，/>representing a significant perceived characteristic of the fluid level->Is a feature vector, ++>Is the number of features;

establishing an initial target tracking model according to the liquid level significant perception characteristics, and carrying out model parameter updating on the initial target tracking model according to a preset parameter updating function to obtain an injection dose target tracking model, wherein the parameter updating function is used for updating parameters of the initial target tracking modelThe function is: ，/>An injection dose target tracking model representing the current moment, < + >>An initial object tracking model representing the previous moment, < +.>Is a smoothing parameter +.>Representing the liquid level significant perception feature, the injection dose target tracking model comprises a two-layer convolution long short-time memory network, a pooling layer and a Sigmiod function.

3. The method of image-based injection dose analysis of claim 1, wherein the performing multi-layer feature recognition and multi-layer feature fusion on the sequence of target injection level images results in a plurality of injection level feature images, comprising:

performing HOG feature extraction on a plurality of image frames in the target injection liquid level image sequence through a preset HOG feature function to obtain HOG features of each image frame, wherein the HOG feature function is as follows:

h represents HOG characteristics, < >>Is the gradient in the x-direction, +.>Is the gradient in the y direction;

respectively toShallow feature extraction is carried out on a plurality of image frames in the target injection liquid level image sequence, so that a shallow feature average value of each image frame is obtained, and the shallow features are as follows:，/>is a single shallow feature, +.>The average value of the shallow features is N, and the number of the shallow features is N;

respectively extracting middle layer characteristics of a plurality of image frames in the target injection liquid level image sequence to obtain middle layer characteristics of each image frame;

Respectively acquiring a first weight parameter of the HOG feature, a second weight parameter of the average value of the shallow features and a third weight parameter of the middle layer feature;

according to the first weight parameter, the second weight parameter and the third weight parameter, carrying out feature fusion calculation on the HOG feature, the shallow feature average value and the middle layer feature through a preset fusion feature function to obtain a target fusion feature, wherein the fusion feature function is as follows:，/>representing the target fusion feature->Represents a first weight parameter, H represents HOG characteristics,>representing a second weight parameter,/->Is the average value of shallow features, +.>Representing a third weight parameter, ++>Representing middle layer characteristics;

and carrying out feature processing on a plurality of image frames in the target injection liquid level image sequence according to the target fusion feature to obtain a plurality of injection liquid level feature images.

4. The image-based injection dose analysis method according to claim 1, wherein the performing saliency region identification on the plurality of injection liquid level feature images respectively to obtain a liquid level saliency region of each injection liquid level feature image, performing feature weight distribution calculation on the liquid level saliency region to obtain a salient feature weight of each injection liquid level feature image, and creating a first injection dose tracking strategy of the injection dose target tracking model according to the salient feature weights comprises:

Performing salient region identification on the injection liquid level characteristic images respectively to obtain a liquid level salient region of each injection liquid level characteristic image;

performing significant perception weight calculation on the liquid level significant region through a preset significant perception weight function to obtain significant perception weight, wherein the significant perception weight function is as follows:，/>representing significant perceptual weights, ++>Representing the distance of the feature point to the salient center;

based on the significant perceived weight, for the liquid levelThe saliency area is subjected to feature adjustment to obtain an adjusted feature, wherein the adjusted feature calculation formula is as follows:，/>the characteristics after the adjustment are represented,representing significant perceptual weights, ++>Representing the original features;

carrying out weight normalization on the adjusted features through a preset weight normalization function to obtain the significant feature weight of each injection liquid level feature image, wherein the weight normalization function is as follows:，/>representing salient feature weights, ++>Representing the adjusted feature;

and setting an influence coefficient of the adjusted feature in the injection dose target tracking model according to the significant feature weight, and generating a corresponding first injection dose tracking strategy according to the influence coefficient.

5. The method of image-based injection dose analysis of claim 1, wherein performing a canonical weight calculation on the plurality of injection level feature images to obtain canonical weight data, and calculating a correlation metric and an actual level tracking position of the plurality of injection level feature images from the canonical weight data comprises:

based on a preset regular weight function, carrying out regular weight calculation on two adjacent image frames in the plurality of injection liquid level characteristic images to obtain regular weight data,，/>representing canonical weight data reflecting the consistency of target level positions of two adjacent image frames, +.>Adjustment coefficient representing position difference, +.>Target level position representing the video frame of the previous image, is->Representing a target liquid level position of a current image video frame;

calculating correlation metrics of the plurality of injection liquid level characteristic images according to the regular weight data, wherein the correlation metric function is as follows:，/>feature vector representing the video frame of the previous image, is->Feature vector representing the video frame of the current image, +.>Representing the association metric +.>Representing canonical weight data，/>An adjustment coefficient representing the characteristic difference;

calculating actual liquid level tracking positions of the plurality of injection liquid level characteristic images according to the correlation metrics by a preset position calculation function, wherein the position calculation function is as follows: ，/>Target level position representing the video frame of the previous image, is->Target level position representing the current image video frame, is->Is a factor of the regulation and is used for regulating the quantity of the liquid,representing a correlation measure between two adjacent frames, +.>Representing the actual level tracking position.

6. The method of image-based injection dose analysis according to claim 1, wherein said performing a consistent response and a response difference calculation on the actual level tracking position according to the injection dose target tracking model to obtain response difference data, and performing a policy adjustment on the first injection dose tracking policy according to the response difference data to obtain a second injection dose tracking policy, comprises:

performing liquid level tracking position prediction on the plurality of injection liquid level characteristic images according to the injection dose target tracking model to obtain a target liquid level tracking position;

based on a preset consistency response function, carrying out consistency response calculation on the actual liquid level tracking position and the target liquid level tracking position to obtain a target consistency response index, wherein the consistency response function is as follows:

，/>is the actual level tracking position, +.>Is the target liquid level tracking position,factor representing the steepness of the control response curve +. >Representing a target consistency response index;

acquiring an actual consistency response index of the injection dose target tracking model, and calculating response difference data of the actual consistency response index and the target consistency response index through a preset response difference function, wherein the response difference function is as follows:，/>representing response difference data, ++>Indicating that the target is consistent with the response index,representing an actual consistency response index;

and carrying out strategy adjustment on the first injection dose tracking strategy according to the response difference data to obtain a second injection dose tracking strategy.

7. The method of claim 1, wherein the performing environmental change evaluation on the target injection level image sequence to obtain an environmental change evaluation index, and performing parameter dynamic adjustment on the second injection dose tracking strategy according to the environmental change evaluation index to obtain a target injection dose tracking strategy of the injection dose target tracking model comprises:

performing environmental change evaluation on the target injection liquid level image sequence through a preset environmental change evaluation function to obtain an environmental change evaluation index, wherein the environmental change evaluation function is as follows: ，/>Indicating an environmental change assessment indicator->Weight factor representing the change of the evaluation environment, +.>Feature vector representing the video frame of the previous image, is->A feature vector representing a video frame of the current image;

carrying out parameter dynamic adjustment calculation on the second injection dose tracking strategy according to the environmental change evaluation index through a preset parameter dynamic adjustment function to obtain a target dynamic tracking parameter, wherein the parameter dynamic adjustment function is as follows:,/>indicating an environmental change assessment indicator->Target level position representing the video frame of the previous image, is->Target level position representing the current image video frame, is->Coefficients representing the adjustment of the tracking parameters +.>Representing a dynamic tracking parameter of the target;

and carrying out parameter dynamic adjustment on the second injection dose tracking strategy according to the target dynamic tracking parameters to obtain a target injection dose tracking strategy of the injection dose target tracking model.

8. An image-based injection dose analysis system, the image-based injection dose analysis system comprising: