Method for transmitting real-time picture of ultrasonic equipment

Abstract

The invention relates to the technical field of image processing, in particular to a method for transmitting real-time pictures of ultrasonic equipment, which comprises the following steps of: acquiring an ultrasonic image through the ultrasonic equipment, dividing the acquired image into block images, and analyzing each block image by using a spatial information analysis model; the method comprises the following steps: identifying a key region and a non-key region in an image through a deep learning model, setting different compression strategies, performing time sequence analysis on acquired bandwidth data by using a bandwidth prediction model, and adjusting a compression ratio, a resolution ratio and a frame rate of the images of the key region and the non-key region according to a prediction result. According to the method, compressed image data is split into a plurality of data streams, a path with high transmission quality is preferentially selected for transmission, multi-stage reconstruction is performed on ultrasonic image data at a receiving end, an image of a key area is processed by using a super-resolution reconstruction algorithm, and high-frequency noise is removed by using a hierarchical denoising algorithm and a convolutional neural network. According to the invention, the real-time performance and reliability of image transmission are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and more particularly to a method for transmitting real-time images of ultrasonic equipment. Background Art

[0002] Ultrasonic medical imaging technology has the advantages of safety, accuracy, no radiation, and easy operation, and has become an important application scenario and node technology for telemedicine. Ultrasonic imaging has dynamic properties and places extremely high demands on the real-time, stability, and bandwidth of data transmission, resulting in technical challenges in the implementation of remote ultrasound services.

[0003] In order to alleviate the huge bandwidth pressure, image playback lag, and asynchrony caused by technical barriers in the process of remote ultrasound image transmission, previous technical methods have adopted measures such as block image comparison, frame rate reduction, image quality reduction, and image resolution reduction, which have improved the efficiency and stability of remote ultrasound image transmission to a certain extent. However, these measures also bring side effects, that is, while improving the transmission efficiency, the image quality, frame rate, and resolution cannot reach the optimal state at the same time. In addition, due to the limitation of network bandwidth, remote ultrasound images need to take into account the transmission of real-time voice and camera image data while transmitting image data, which undoubtedly further weakens the quality and stability of remote ultrasound real-time image transmission.

[0004] Based on the above problems, announcement number CN116705280 A discloses a remote transmission control system and method for ultrasound images. This technology divides ultrasound images into regions of interest and remaining regions, and dynamically decides to use lossless compression, lossy compression, or original quality transmission based on the health status of the network to ensure that ultrasound images can still be effectively transmitted when the network is restricted. By dynamically adjusting transmission parameters based on network status and image characteristics, the system can transmit high-quality images that meet diagnostic requirements under different network conditions, ensuring the continuity and stability of telemedicine services.

[0005] The above-mentioned existing public technologies dynamically adjust the compression ratio, resolution and frame rate of the image when the network bandwidth fluctuates, resulting in unclear differences in transmission quality between key areas and non-key areas, making it difficult to optimize image clarity and transmission speed under limited bandwidth. Summary of the invention

[0006] In view of the shortcomings of the prior art, the object of the present invention is to provide a method for transmitting real-time images of ultrasound equipment with bandwidth prediction, adaptive compression, multi-path dynamic scheduling and priority control.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A method for transmitting real-time images of an ultrasonic device comprises the following steps:

[0009] Step S1: Acquire an ultrasonic image in real time through an ultrasonic device, and divide the acquired image into a number of irregularly shaped block images.

[0010] The spatial information analysis model is used to analyze the spatial continuity between each block image, including calculating the relative position, similarity and edge connectivity between each block image.

[0011] During the analysis process, the spatial information between adjacent block images is associated according to the texture features and edge features of the block images, and the position of each block image is marked for subsequent compression and transmission processing;

[0012] Step S2: Identify the key areas and non-key areas in the image through the deep learning model and set different compression strategies.

[0013] Lossless compression is used in key areas, and full-fidelity compression is performed on the pixels of each block image in the key area to ensure lossless image details; lossy compression is used in non-key areas, and the compression rate is dynamically adjusted based on the texture complexity of key and non-key areas.

[0014] The initial data sequence is compressed layer by layer using an adaptive compression algorithm. For consecutive pixels with the same characteristics, a symbol replacement model is used to convert the continuous data into shorter symbol codes to reduce data redundancy.

[0015] During the compression process, the compression path and replacement symbol information of each block image is recorded as additional information of the compressed package to ensure rapid restoration during decompression;

[0016] Step S3: Use network monitoring tools to collect real-time network bandwidth status, including current bandwidth, latency, packet loss rate and other performance parameters.

[0017] Use bandwidth prediction models to perform time series analysis on the collected bandwidth data and predict future bandwidth fluctuations based on historical bandwidth change trends.

[0018] According to the prediction results, the compression ratio, resolution and frame rate of the images in the key and non-key areas are adaptively adjusted through the adjustment module.

[0019] When bandwidth is sufficient, use high resolution and low compression ratio transmission for key areas.

[0020] When bandwidth is insufficient, the resolution and frame rate are reduced for non-critical areas to save bandwidth;

[0021] When the network status fluctuates, the transmission frame rate and data flow priority of each area are dynamically adjusted to ensure stable picture quality in key areas.

[0022] Step S4: split the compressed image data into multiple data streams, and preferentially select a path with high transmission quality for transmission according to the importance of the image content and the load of the network path;

[0023] Set up a fault-tolerant mechanism for multi-path transmission. If a bandwidth fluctuation or excessive delay is detected on a transmission path, it will automatically switch to other stable paths to prevent data interruption caused by a single path failure.

[0024] Redundantly encode the data in key areas and send them to multiple paths to ensure that in the event of data loss, data from other paths can be used for recovery, thereby improving image integrity.

[0025] Step S5: At the receiving end, the received ultrasound image data is reconstructed in multiple levels, with the detailed information of the key areas being reconstructed first, and then the non-key areas are loaded step by step.

[0026] Use super-resolution reconstruction algorithms to magnify the details of key areas of the image and restore the image clarity lost due to compression.

[0027] A hierarchical denoising algorithm is used, first using median filtering to remove low-frequency noise in the compression process, and then using a convolutional neural network to remove high-frequency noise to ensure image clarity and diagnostic quality.

[0028] The present invention is further configured as follows: the analysis steps of the spatial information analysis model in step S1 are as follows:

[0029] Similarity calculation between block images: Assume that each block image is , each block image contains pixels, represented as a matrix , for adjacent block images and , using the similarity calculation formula based on cosine:

[0030]

[0031] Similarity value The closer it is to 1, the more similar the contents of the two block images are. You can set the threshold ,when > When , it is regarded as a block image and Continuous in space;

[0032] For each block image and , calculate the average gray value of its edge pixels respectively:

[0033]

[0034] in, Represents a block image The edge pixel set, N is the number of edge pixels, if < ( is the set connectivity threshold), then the edges of the two block images are considered to be connected.

[0035] When performing spatial analysis on the block images, in order to retain the position of each block image in the original image, a position encoding mechanism is introduced to enhance the accuracy of spatial information by encoding the position coordinates. Specifically:

[0036] Assuming the image is segmented The coordinates of , sine and cosine functions encode the position and generate a position encoding vector ,

[0037]

[0038] Where d represents the dimension of the position code, and k is the dimension index of the position code vector. The code can be directly added to the feature vector of the block image, so that the spatial information of the block image can be retained in subsequent calculations.

[0039] After completing the spatial continuity analysis and position encoding, the model needs to prioritize the block images and set the sorting weights. , and Respectively represent the influence of these three factors. For the block image , its priority weight The calculation formula is:

[0040]

[0041] According to weight The size of the image blocks is prioritized, and the image blocks with larger weights are marked as important areas first. Subsequently, a lossless compression ratio is used to ensure spatial continuity and visual effects.

[0042] The present invention is further configured as follows: in step S2, the CNN model is used to analyze the image, and the diagnostic importance of each block image is predicted, assuming that the block image The importance rating is , the output layer of the CNN model can predict the rating:

[0043]

[0044] score Above threshold The block images of are marked as key areas, and the score is lower than The area marked as non-critical;

[0045] For key area segmented images Set a low compression ratio ; For non-critical area block images Set a high compression ratio , based on the importance of the region With the set minimum and maximum compression ratio Calculate the compression ratio for each region:

[0046]

[0047] in is the maximum value of the importance score, where Minimum importance score.

[0048] The present invention is further configured as follows: the symbol replacement model is used to replace repeated data with shorter symbols to reduce the amount of data, and the symbol replacement includes: for each block image A sequence of consecutive pixels in , if the frequency of occurrence of the sequence reaches the set threshold, it will be marked with the symbol Instead, that is = , replace the symbol table Stored in the compressed information, the symbol replacement table is used to restore the original data during decompression.

[0049] For the symbols generated by the current compression stage , record its position in the data sequence and frequency , the update formula of the compression information table is After compression is completed, the compression information table contains the replacement information and position index of all symbols to ensure that the data can be correctly restored during transmission and decompression;

[0050] For data in non-critical areas, differential encoding is further used to reduce the amount of information, and pixel values ​​with no significant differences are represented by differential symbols to reduce storage occupancy.

[0051] The present invention is further configured as follows: the model of the compression ratio in step 3 is:

[0052]

[0053]

[0054] in, is the maximum bandwidth value; is the minimum compression ratio; is the maximum compression ratio; is the compression ratio of the key area; is the compression ratio of non-critical areas; is the bandwidth prediction value.

[0055] The present invention is further configured as follows: the resolution adjustment model in step 3 is:

[0056]

[0057]

[0058] is the maximum resolution, is the resolution of non-critical areas, is the resolution of the critical area, To control the resolution sensitivity parameters and adapt to the change of bandwidth,

[0059] The frame rate adjustment model is:

[0060]

[0061]

[0062] is the maximum frame rate, is the frame rate of the non-critical area, is the frame rate of the key area, Adjust the sensitivity factor for frame rate.

[0063] The present invention is further configured as follows: before the receiving end in step S5 receives the data transmission, the transmitting end calculates the transmission order of the key area and the non-key area according to the dynamic priority weight, so as to set the loading priority for different layers of the key area and the non-key area under different bandwidth conditions, and the layer with a higher priority weight will be transmitted and loaded first under the condition of tight bandwidth.

[0064] The dynamic priority weight formula is as follows:

[0065]

[0066] and They are the dynamic priority weights of the i-th layer in the critical area and the non-critical area, ranging from 0 to 1. The higher the weight value, the higher the priority loading demand of the layer;

[0067] and They are the initial priority coefficients of the critical area and the non-critical area at the i-th layer, usually > To ensure priority in key areas,

[0068] and The weighted sum of the priorities of all critical area and non-critical area layers is used for normalization calculation. The denominator of the sum ensures that the calculated weight is between 0 and 1.

[0069] The present invention is further configured to dynamically adjust the loading depth according to the current available bandwidth through an adaptive hierarchical loading model, wherein:

[0070]

[0071] in, is the maximum number of layers of the key area that the current network bandwidth allows to load; L is the total number of layers of the key area; i is an iteration variable, which is used to indicate the order of layer-by-layer loading; is the cumulative quality increment, i.e., the quality value of the first i image detail layers after accumulation; is the current available bandwidth; T is the minimum threshold for the key area to meet the basic quality requirements.

[0072] The present invention is further configured as follows: step S4 also includes real-time path detection, real-time monitoring of available network paths, collecting load information of each path, and calculating the transmission quality score of each path according to the load condition of the path;

[0073] Prioritize the allocation of high-scoring paths to data flows in key areas, ensuring that data in key areas has more network resources when being transmitted;

[0074] Assign priorities to data streams based on the importance of image content, and bind high-priority data streams to paths with higher scores;

[0075] After the data flow in the key area is transmitted stably, the data flow in the non-key area will be allocated to the remaining paths, and its transmission quality will be gradually increased when the bandwidth allows.

[0076] When a path becomes overloaded or fails, the data stream bound to the path will be automatically switched to other paths with higher scores to ensure uninterrupted transmission.

[0077] The present invention is further configured as follows: a path scoring model:

[0078]

[0079] is the score of path p; is the path bandwidth; is the path delay; is the packet loss rate; , , are the weights of bandwidth, delay, and packet loss rate, respectively; , , is the maximum bandwidth, minimum delay and minimum packet loss rate of the available path, which is used for normalization.

[0080] Based on the score value, bind the data flow to the path as follows:

[0081]

[0082] in Indicates the preferred path for data flow in key areas. If the status of this path changes, the system will recalculate the score and automatically switch to the next optimal path.

[0083] Compared with the shortcomings of the prior art, the beneficial effects of the present invention are:

[0084] This application adopts a bandwidth prediction and adaptive adjustment mechanism to dynamically adjust the compression ratio, resolution and frame rate of the image according to the network status, so that high-resolution and low-compression ratio image data can be transmitted when the bandwidth is sufficient, and when the bandwidth is insufficient, the bandwidth can be saved by reducing the transmission quality of non-critical areas, thereby ensuring the transmission stability of critical areas.

[0085] Combined with the real-time network load situation, high-quality transmission paths are prioritized for data streams in key areas, and dynamic switching is performed when the load is too high, realizing intelligent path selection and load balancing, reducing transmission delays and packet loss rates in key areas, and further improving the real-time and reliability of image transmission. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 It is a schematic diagram of the process of the present invention;

[0087] Figure 2 Analogous diagram optimized for the present invention and two existing image formats. DETAILED DESCRIPTION

[0088] Reference Figure 1 to Figure 2 The embodiments of the present invention are further described.

[0089] A method for transmitting real-time images of ultrasonic equipment,

[0090] Step S1: The ultrasound device collects and transmits image frame data to the processing module in real time. This process ensures that the diagnostic information in each frame is complete and synchronized. The image is divided into several irregularly shaped block images to facilitate distinguishing key areas (such as lesion areas or other parts rich in diagnostic information) from non-key areas, so that key areas can be identified and prioritized to ensure that higher priority is given to key areas in subsequent transmission and compression. This is set according to the edge features and texture changes of the image, which helps to focus on key areas. At the same time, a lower priority is designed for non-key areas to reduce the burden of data transmission.

[0091] In order to determine the similarity between the block images, the system converts each block image into , considered as containing pixels, represented as a matrix ,in Represents a block image The j-th pixel value in the adjacent block image and , similarity is calculated using a formula based on cosine similarity:

[0092]

[0093] Similarity value The closer it is to 1, the more similar the contents of the two block images are. Set the threshold ,when > When , it is regarded as a block image and Continuous in space;

[0094] For each block image and , calculate the average gray value of its edge pixels to determine the edge connection. The specific formula is:

[0095]

[0096] in, Represents a block image The edge pixel set, N is the number of edge pixels, if < ( is the connectivity threshold set), then the edges of the two block images are considered to be connected.

[0097] Similarity calculation and edge connectivity calculation help identify connected areas in the image and treat block images with similar texture or edge features as continuous. This ensures that the spatial structure of the image will not be destroyed due to block during reconstruction and transmission.

[0098] When performing spatial analysis on the block images, in order to retain the position of each block image in the original image, a position encoding mechanism is introduced to enhance the accuracy of spatial information by encoding the position coordinates. The coordinates of , use sine and cosine functions to encode the position and generate a position encoding vector ,

[0099]

[0100] Among them, d represents the dimension of the position code, k is the dimension index of the position code vector, and the code can be directly added to the feature vector of the block image so that the spatial information of the block image can be retained in subsequent calculations.

[0101] The introduction of position coding further improves the spatial recognition ability of the block image. By encoding the coordinates of each block image, it ensures that each image block maintains its unique spatial position during analysis and transmission. Position coding combined with other spatial analysis can enhance the spatial consistency and integrity of the image in subsequent steps.

[0102] After completing the spatial continuity analysis and position encoding, the model needs to prioritize the block images and set the sorting weights. , and Respectively represent the influence of these three factors. For the block image , its priority weight The calculation formula is:

[0103]

[0104] According to weight The size of the image blocks is prioritized, and the image blocks with larger weights are marked as important areas first. Subsequently, a lossless compression ratio is used to ensure spatial continuity and visual effects.

[0105] Priority sorting lays the foundation for subsequent data compression and transmission strategies. By analyzing the similarity, edge connectivity and spatial position of each block image, the block images with higher weights are marked as important areas. These important areas will be given priority to lossless compression during data transmission to ensure clear details, while blocks with lower weights can use lossy compression to reduce the amount of data.

[0106] This hierarchical processing method optimizes data transmission efficiency and can provide support for real-time diagnosis by giving priority to the transmission of high-weight image data in a network environment with limited bandwidth, thereby achieving fast and high-definition presentation of key area information.

[0107] Step S2: Use the convolutional neural network (CNN) model to perform content analysis on the acquired ultrasound images. The CNN model can extract multi-level features of the image and analyze each block image layer by layer to predict the diagnostic importance of each block. Assume that each block image is , the output layer of the CNN model generates an importance score based on the input image features , the formula is as follows:

[0108]

[0109] In this model, the score Represents a block image The CNN model learns the diagnostic features of different image regions through training data and can effectively distinguish between key and non-key regions. The score is higher than the set threshold. The block images of are marked as key areas, while those below The image is marked as non-critical area;

[0110] The image scores of the blocks predicted by the CNN model Different compression strategies are assigned to key areas and non-key areas. To ensure the image quality of key areas, a low compression ratio is set for key areas. , and set a high compression ratio for non-critical areas

[0111] The compression ratio is calculated as follows:

[0112] is the compression ratio of the block image, are the minimum and maximum values ​​of the compression ratio,

[0113] in is the maximum value of the importance score, where Minimum importance score.

[0114] This formula ensures that the image blocks with high importance scores (i.e., key areas) use a lower compression ratio to maintain high quality, while for areas with lower scores (i.e., non-key areas), a higher compression ratio is used to reduce the amount of data and save bandwidth.

[0115] In order to maintain diagnostic details in key areas, the system uses lossless compression for these areas. This method compresses each pixel in the block image with full fidelity to ensure that no image details are lost. Lossless compression can be achieved using common compression algorithms (such as PNG or WebP), and the specific algorithm is selected based on image characteristics and compression requirements.

[0116] For non-critical areas, the system uses a lossy compression strategy to reduce the amount of data. According to the texture complexity of the area, the system will also dynamically adjust the compression rate to ensure that non-critical areas occupy as little space as possible when bandwidth is limited.

[0117] The initial data sequence in non-critical areas is compressed layer by layer through an adaptive compression algorithm. This algorithm identifies similar or repeated features and uses a symbol replacement model to convert continuous identical data into short symbol codes, thereby effectively reducing data redundancy.

[0118] The symbol replacement model is used to replace repeated data with shorter symbols during the compression process to reduce the amount of data. Assume that each block image Contains a set of continuous pixel sequences , if the frequency of occurrence of the sequence reaches the set threshold, it will be marked with the symbol Instead, that is = , replace the symbol table Stored in the compressed information, it is used to restore the original data when decompressing. This replacement method can significantly reduce redundant data and improve compression efficiency.

[0119] During the compression process, the system records the compression path and replacement symbol information of each block image and generates a compression information table. The table contains the symbol replacement information, position index, and the frequency of occurrence of the symbol in the data sequence. The update formula of the compression information table is:

[0120] in is the symbol, is the symbol position, is the frequency of occurrence of the symbol. After compression is completed, the compression information table ensures that the data can be correctly restored during transmission and decompression.

[0121] For data in non-critical areas, differential encoding is further used to reduce the amount of information, and pixel values ​​with no significant differences are represented by differential symbols to reduce storage occupancy.

[0122] Through this method, the system replaces pixel values ​​with differential symbols without significant differences, avoiding the storage of duplicate data and further reducing the amount of data.

[0123] Step S3: Use network monitoring tools to collect network bandwidth status parameters in real time, including the current bandwidth , Network delay and packet loss rate The network bandwidth monitoring system can capture dynamically changing data within millisecond time intervals and record bandwidth fluctuations in real time.

[0124] The bandwidth prediction model uses time series analysis to record historical bandwidth data. Predict future bandwidth trends to determine appropriate compression and transmission strategies.

[0125] Specific formula of prediction model:

[0126] in To predict the future bandwidth value, the input is the historical bandwidth data. The bandwidth prediction model usually adopts the LSTM (Long Short-Term Memory Network) structure, which can learn the long-term dependencies in the time series and predict the bandwidth fluctuation trend.

[0127] The purpose of bandwidth prediction is to determine the bandwidth change trend in advance and provide the system with a stable bandwidth reference value so that timely adjustments can be made when the bandwidth is about to be insufficient or sufficient to ensure efficient image transmission.

[0128] According to the prediction results, the adjustment module adaptively adjusts the compression ratio, resolution and frame rate of the image. The adjustment process is based on the real-time monitoring of bandwidth status and its predicted trend to ensure high-quality transmission in key areas. The specific adjustment strategy is as follows:

[0129] Compression ratio adjustment: When bandwidth is sufficient, low compression ratio is used in key areas When bandwidth is insufficient, the compression ratio of non-critical areas is increased to To save bandwidth resources. The dynamic adjustment model of compression ratio is as follows:

[0130]

[0131]

[0132] in, is the maximum bandwidth value; is the minimum compression ratio; is the maximum compression ratio; is the compression ratio of the key area; is the compression ratio of non-critical areas; is the bandwidth prediction value.

[0133] Resolution adjustment: The system dynamically adjusts the image resolution based on bandwidth prediction results. For key areas, when bandwidth is sufficient, the highest resolution is used. When the bandwidth is insufficient, the resolution of non-critical areas is reduced to , the resolution adjustment formula is as follows:

[0134]

[0135]

[0136] is the maximum resolution, is the resolution of non-critical areas, is the resolution of the critical area, The parameter that controls resolution sensitivity determines how strongly the resolution responds to changes in bandwidth. This adjustment ensures that resolution in critical areas is prioritized when bandwidth is limited.

[0137] Frame rate adjustment: Based on the results of bandwidth prediction, the frame rate will be adjusted differently between key areas and non-key areas. The goal of frame rate adjustment is to maintain the smoothness of the picture in key areas. The frame rate adjustment formula is:

[0138]

[0139]

[0140] is the maximum frame rate, is the frame rate of the non-critical area, is the frame rate of the key area, Adjusts the sensitivity factor for frame rate. This formula ensures that critical areas use the maximum frame rate when bandwidth is sufficient, while non-critical areas gradually reduce the frame rate when bandwidth is insufficient.

[0141] When the network status fluctuates, the transmission frame rate and data flow are dynamically adjusted according to the priority of the image area to ensure that the key areas maintain stable image quality during fluctuations. ,Each block image is scored according to parameters such as the image’s diagnostic importance and transmission priority, ensuring that high-priority areas maintain high fluency when the network fluctuates.

[0142] Step S4: First, the compressed image data is split into multiple data streams, and the best transmission path is allocated to different data streams according to the importance of each block image content (such as key area and non-key area) and the load of the network path.

[0143] Each data stream is prioritized based on the importance of the image content, and multi-path selection is performed through a path scoring model. The multi-path transmission design ensures that under high load or network fluctuations, data streams can be transmitted simultaneously on multiple paths to prevent interruptions in single path transmission.

[0144] To select a suitable transmission path, the system designs a path scoring model that takes into account the bandwidth of each path. ,Delay , Packet loss rate The transmission quality of each path is calculated using the following scoring formula:

[0145]

[0146] is the score of path p; is the path bandwidth; is the path delay; is the packet loss rate; , , are the weights of bandwidth, delay and packet loss rate respectively; , , is the maximum bandwidth, minimum delay and minimum packet loss rate of the available path, which is used for normalization.

[0147] Fault-tolerance mechanism and dynamic path switching: To ensure the stability of data transmission, a fault-tolerance mechanism is set on each path. If bandwidth fluctuations or excessive delays are detected on a transmission path, it will automatically switch to a path with a higher score to prevent transmission interruptions caused by a single path failure.

[0148] For data in critical areas, redundant coding (such as error correction code or repeated transmission mechanism) is applied to send data to multiple paths to ensure that even if packet loss occurs on one path, data from other paths can still be used for data recovery, thereby improving image integrity.

[0149] Real-time path detection and load monitoring: All available network paths are monitored in real time, and the load information of each path is collected, including the path's current bandwidth, latency, packet loss rate and other indicators. These real-time monitoring data are used to update the path scoring model to ensure that each path allocation is based on the latest network status information.

[0150] The real-time path detection mechanism enables the system to dynamically respond to changes in network status, quickly adjust data flows and reallocate paths to improve transmission stability and efficiency.

[0151] Data flow priority allocation and path binding:

[0152] Assign a priority to each data stream based on the importance of the image content (critical area and non-critical area), and bind high-priority data streams to paths with higher scores. Path score value Higher paths are assigned priority to key area data flows to ensure that critical data receives priority network resources during transmission.

[0153] The formula is as follows:

[0154] in Indicates the preferred path for data flow in key areas. If the status of this path changes, the system will recalculate the score and automatically switch to the next optimal path to ensure transmission quality.

[0155] Dynamic priority weight calculation: At the transmission end, the transmission order of data streams in key areas and non-key areas is set according to the dynamic priority weight, so as to set loading priorities for different layers of key areas and non-key areas under different bandwidth conditions. The layers with higher priority weights will be transmitted and loaded first when bandwidth is tight.

[0156] The dynamic priority weight calculation formula is as follows:

[0157]

[0158] and They are the dynamic priority weights of the i-th layer in the critical area and the non-critical area, ranging from 0 to 1. The higher the weight value, the higher the priority loading demand of the layer;

[0159] and They are the initial priority coefficients of the critical area and the non-critical area at the i-th layer, usually > To ensure priority in key areas,

[0160] and The weighted sum of the priorities of all critical area and non-critical area layers is used for normalization calculation. The denominator of the sum ensures that the calculated weight is between 0 and 1.

[0161] Through the design of multi-path transmission and fault-tolerant mechanisms, combined with the adaptive adjustment of dynamic priority weights, priority transmission and quality assurance of data in key areas are ensured, effectively improving the system's transmission efficiency and image integrity in complex network environments.

[0162] Step S5: At the receiving end, the received ultrasound image data is first reconstructed at multiple levels. The design of multi-level reconstruction is based on the importance of the image and the available bandwidth, and the details of the key areas are loaded first to ensure that the doctor can quickly obtain key diagnostic information at the receiving end. Specifically, the image details are divided into multiple levels, and the key areas are reconstructed with the highest resolution and detail level first, while non-key areas are loaded step by step according to the network conditions. This method effectively improves the response speed of the image, especially under conditions of limited bandwidth, and the core details of the key areas can be displayed as soon as possible.

[0163] The system uses an adaptive hierarchical loading model to load Dynamically adjust the image loading level depth, the adaptive layered loading model:

[0164]

[0165] in, is the maximum number of layers of the key area that the current network bandwidth allows to load; L is the total number of layers of the key area; i is an iteration variable, which is used to indicate the order of layer-by-layer loading; is the cumulative quality increment, i.e., the quality value of the first i image detail layers after accumulation; is the current available bandwidth; T is the minimum threshold for the key area to meet the basic quality requirements.

[0166] The formula adaptively adjusts the number of loaded layers through dynamic evaluation of the current bandwidth, ensuring that more layers of image details can be loaded as quickly as possible when the bandwidth is sufficient; when the bandwidth is insufficient, the basic layer is loaded first to meet basic diagnostic needs.

[0167] In order to restore image details lost due to compression, the system uses a super-resolution reconstruction algorithm to magnify image details in key areas. The algorithm uses a deep neural network (DNN) to magnify low-resolution images, restore details and enhance image clarity.

[0168] The input of the super-resolution reconstruction algorithm is the received compressed image block, and the output is the enlarged high-resolution image. This process can learn the high-frequency details of the ultrasound image through the generative adversarial network (GAN) structure of the convolutional neural network (CNN), and effectively remove blur when enlarging the image to ensure the diagnostic quality of key areas.

[0169] During the reconstruction process, the system uses a hierarchical denoising algorithm to process the image noise to eliminate the low-frequency and high-frequency noise generated during the transmission and compression process to ensure the diagnostic quality of the final image.

[0170] Hierarchical denoising includes the following two steps:

[0171] Median filtering: First, median filtering is performed on the image to remove low-frequency noise generated during the transmission process. Median filtering has a good effect on image edge preservation, so it is suitable for preserving structural information in key areas.

[0172] High-frequency denoising: After removing low-frequency noise, a convolutional neural network (CNN) is used to process the high-frequency noise of the image to further improve the clarity of the image. The high-frequency denoising process can specifically remove subtle high-frequency noise, such as artifacts or small image errors.

[0173] Guarantee of image clarity and diagnostic quality:

[0174] Through multi-level reconstruction, super-resolution reconstruction and hierarchical denoising, the system effectively restores image details that may be lost during compression transmission at the receiving end, ensuring the clarity and stability of key areas. This combined approach improves the diagnostic quality of images, allowing doctors to accurately observe key details and improve the reliability of diagnosis.

[0175] like Figure 2 This is a data chart of PSNR and bit rate between the present technology and JPEG and Webp. As can be seen from the chart, the present technology has a higher PSNR value at a low bit rate, that is, the image quality is higher than that of traditional JPEG and Webp.

[0176] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Common changes and substitutions made by those skilled in the art within the scope of the technical solution of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for transmitting real-time images of ultrasonic equipment, characterized in that: The steps include: Step S1: Acquire an ultrasonic image in real time through an ultrasonic device, and divide the acquired image into a number of irregularly shaped block images. The spatial information analysis model is used to analyze the spatial continuity between each block image, including calculating the relative position, similarity and edge connectivity between each block image. During the analysis process, the spatial information between adjacent block images is associated according to the texture features and edge features of the block images, and the position of each block image is marked for subsequent compression and transmission processing; Step S2: Identify the key areas and non-key areas in the image through the deep learning model and set different compression strategies. Lossless compression is used in key areas, and full-fidelity compression is performed on the pixels of each block image in the key area to ensure lossless image details; lossy compression is used in non-key areas, and the compression rate is dynamically adjusted based on the texture complexity of key and non-key areas. The initial data sequence is compressed layer by layer using an adaptive compression algorithm. For consecutive pixels with the same characteristics, a symbol replacement model is used to convert the continuous data into shorter symbol codes to reduce data redundancy. During the compression process, the compression path and replacement symbol information of each block image is recorded as additional information of the compressed package to ensure rapid restoration during decompression; Step S3: Use network monitoring tools to collect real-time network bandwidth status, including current bandwidth, latency, packet loss rate and other performance parameters. Use bandwidth prediction models to perform time series analysis on the collected bandwidth data and predict future bandwidth fluctuations based on historical bandwidth change trends. According to the prediction results, the compression ratio, resolution and frame rate of the images in the key and non-key areas are adaptively adjusted through the adjustment module. When bandwidth is sufficient, use high resolution and low compression ratio transmission for key areas. When bandwidth is insufficient, the resolution and frame rate are reduced for non-critical areas to save bandwidth. When the network status fluctuates, the transmission frame rate and data flow priority of each area are dynamically adjusted to ensure stable image quality in key areas; Step S4: split the compressed image data into multiple data streams, and preferentially select a path with high transmission quality for transmission according to the importance of the image content and the load of the network path; Set up a fault-tolerant mechanism for multi-path transmission. If a bandwidth fluctuation or excessive delay is detected on a transmission path, it will automatically switch to other stable paths to prevent data interruption caused by a single path failure. Redundantly encode the data in key areas and send them to multiple paths, ensuring that in the event of data packet loss, data from other paths can be used for recovery, thereby improving image integrity; Step S5: At the receiving end, the received ultrasound image data is reconstructed in multiple levels, with the detailed information of the key areas being reconstructed first, and then the non-key areas are loaded step by step. Use super-resolution reconstruction algorithms to magnify the details of key areas of the image and restore the image clarity lost due to compression. A hierarchical denoising algorithm is used, first using median filtering to remove low-frequency noise in the compression process, and then using a convolutional neural network to remove high-frequency noise to ensure image clarity and diagnostic quality.

2. A method for transmitting real-time images of ultrasonic equipment according to claim 1, characterized in that: The analysis steps of the spatial information analysis model in step S1 are as follows: Similarity calculation between block images: Assume that each block image is , each block image contains pixels, represented as a matrix , for adjacent block images and , using the cosine-based similarity formula: Similarity value The closer it is to 1, the more similar the contents of the two block images are. You can set the threshold ,when > When , it is regarded as a block image and Continuous in space; For each block image and , calculate the average gray value of its edge pixels respectively: in, Represents a block image The edge pixel set, N is the number of edge pixels, if < ( is the set connectivity threshold), then the edges of the two block images are considered to be connected. When performing spatial analysis on the block images, in order to retain the position of each block image in the original image, a position encoding mechanism is introduced to enhance the accuracy of spatial information by encoding the position coordinates. Specifically: Assuming the image is segmented The coordinates of , sine and cosine functions encode the position and generate a position encoding vector , Where d represents the dimension of the position code, and k is the dimension index of the position code vector. The code can be directly added to the feature vector of the block image, so that the spatial information of the block image can be retained in subsequent calculations. After completing the spatial continuity analysis and position encoding, the model needs to prioritize the block images and set the sorting weights. , and Respectively represent the influence of these three factors. For the block image , its priority weight The calculation formula is: According to weight The size of the image blocks is prioritized, and the image blocks with larger weights are marked as important areas first. Subsequently, a lossless compression ratio is used to ensure spatial continuity and visual effects.

3. A method for transmitting real-time images of ultrasonic equipment according to claim 2, characterized in that: In step S2, the CNN model is used to analyze the image and predict the diagnostic importance of each block image. Assuming that the block image The importance rating is , the output layer of the CNN model can predict the rating: score Above threshold The block images of are marked as key areas, and the score is lower than The area marked as non-critical; For key area segmented images Set a low compression ratio ; For non-critical area segmentation images Set a high compression ratio , based on the importance of the region With the set minimum and maximum compression ratio Calculate the compression ratio for each region: in is the maximum value of the importance score, where Minimum importance score.

4. A method for transmitting real-time images of ultrasonic equipment according to claim 3, characterized in that: The symbol replacement model is used to replace repeated data with shorter symbols to reduce the amount of data. The symbol replacement includes: for each block image A sequence of consecutive pixels in , if the frequency of occurrence of the sequence reaches the set threshold, it will be marked with the symbol Instead, that is = , replace the symbol table Stored in the compressed information, the symbol replacement table is used to restore the original data during decompression. For the symbols generated by the current compression stage , record its position in the data sequence and frequency , the update formula of the compression information table is After compression is completed, the compression information table contains the replacement information and position index of all symbols to ensure that the data can be correctly restored during transmission and decompression; For data in non-critical areas, differential encoding is further used to reduce the amount of information, and pixel values ​​with no significant differences are represented by differential symbols to reduce storage occupancy.

5. A method for transmitting real-time images of ultrasonic equipment according to claim 4, characterized in that: The model of the compression ratio in step 3 is: in, is the maximum bandwidth value; is the minimum compression ratio; is the maximum compression ratio; is the compression ratio of the key area; is the compression ratio of non-critical areas; is the bandwidth prediction value.

6. A method for transmitting real-time images of ultrasonic equipment according to claim 5, characterized in that: The resolution adjustment model in step 3 is: ； is the maximum resolution, is the resolution of non-critical areas, is the resolution of the critical area, To control the resolution sensitivity parameters and adapt to the change of bandwidth, The frame rate adjustment model is: is the maximum frame rate, is the frame rate of the non-critical area, is the frame rate of the key area, Adjust the sensitivity factor for frame rate.

7. A method for transmitting real-time images of ultrasonic equipment according to claim 6, characterized in that: Before the receiving end receives the data transmission in step S5, the transmitting end calculates the transmission order of the key area and the non-key area according to the dynamic priority weight, so as to set the loading priority for different layers of the key area and the non-key area under different bandwidth conditions. The layer with a higher priority weight will be transmitted and loaded first under the condition of tight bandwidth. The dynamic priority weight formula is as follows: and They are the dynamic priority weights of the i-th layer in the critical area and the non-critical area, ranging from 0 to 1. The higher the weight value, the higher the priority loading demand of the layer; and They are the initial priority coefficients of the critical area and the non-critical area at the i-th layer, usually > To ensure priority in key areas, and The weighted sum of the priorities of all critical area and non-critical area layers is used for normalization calculation. The denominator of the sum ensures that the calculated weight is between 0 and 1.

8. A method for transmitting real-time images of ultrasonic equipment according to claim 7, characterized in that: The loading depth is dynamically adjusted based on the currently available bandwidth through an adaptive hierarchical loading model. The adaptive hierarchical loading model: in, is the maximum number of layers of the key area that the current network bandwidth allows to load; L is the total number of layers of the key area; i is an iteration variable, which is used to indicate the order of loading layer by layer; is the cumulative quality increment, i.e., the quality value of the first i image detail layers after accumulation; is the current available bandwidth; T is the minimum threshold for the key area to meet the basic quality requirements.

9. A method for transmitting real-time images of ultrasonic equipment according to claim 8, characterized in that: Step S4 also includes real-time path detection, which monitors the available network paths in real time, collects the load information of each path, and calculates the transmission quality score of each path according to the load of the path; Prioritize the allocation of high-scoring paths to data flows in key areas, ensuring that data in key areas has more network resources when being transmitted; Assign priorities to data streams based on the importance of image content, and bind high-priority data streams to paths with higher scores; After the data flow in the key area is transmitted stably, the data flow in the non-key area will be allocated to the remaining paths, and its transmission quality will be gradually increased when the bandwidth allows. When a path becomes overloaded or fails, the data stream bound to the path will be automatically switched to other paths with higher scores to ensure uninterrupted transmission.

10. A method for transmitting real-time images of ultrasonic equipment according to claim 9, characterized in that: Model for path scoring: is the score of path p; is the path bandwidth; is the path delay; is the packet loss rate; , , are the weights of bandwidth, delay and packet loss rate respectively; , , is the maximum bandwidth, minimum delay and minimum packet loss rate of the available path, which is used for normalization. Based on the score value, bind the data flow to the path as follows: in Indicates the preferred path for data flow in key areas. If the status of this path changes, the system will recalculate the score and automatically switch to the next optimal path.

Images