Medical instrument intelligent identification method, device and equipment and medium

Abstract

The invention relates to the technical field of computer vision and medical instrument management, and discloses a medical instrument intelligent identification method, device and equipment and a medium for solving the problem that the accuracy of surgical instrument identification is insufficient. And a dual-polarization acquisition and pixel difference technology is combined to accurately separate a mirror surface highlight area from a diffuse reflection area, so that the image definition is improved. Through the first analysis model, the determined and uncertain identification areas are rapidly determined, and the second image of the uncertain area is focused and complementarily collected, so that redundant collection is reduced, and the data efficiency and accuracy are improved. The image subjected to glare-removing tuning processing has clearer feature information, and the recognition accuracy of the uncertain area can be remarkably improved after re-analysis. And finally, fusing the first and second determination identification results to comprehensively improve the accuracy and reliability of medical instrument identification, thereby effectively solving the technical problem of insufficient accuracy of existing medical instrument intelligent identification.

Description

Technical Field

[0001] The present invention relates to the field of computer vision and medical device management technology, and in particular to a medical device intelligent identification method, device, equipment and medium. Background Art

[0002] In the existing closed-loop management process for surgical instruments, medical devices typically follow a circulation path of "issued by the sterilization supply center - used in the operating room - recovered by the sterilization supply center." The specific process includes preoperative preparation in the supply room, the circulation and distribution of sterilization packs, instrument inventory during surgery, and review after recovery by the supply room. Each of these steps relies heavily on manual experience to verify the type, quantity, and integrity of the instruments. This is especially true during the inventory in the operating room and the review of recovery by the supply room, where each instrument must be checked piece by piece. This is tedious and time-consuming, and is prone to risks such as missed or incorrect clearances due to human oversight. In serious cases, this can impact surgical safety and the accuracy of traceability management. With the advancement of image recognition and detection technologies, some medical device management processes have begun to experiment with the introduction of auxiliary recognition solutions based on intelligent analysis models (such as deep neural network models) to enable error correction for omissions, excesses, and placement of medical devices, improving the accuracy and efficiency of instrument inventory and verification.

[0003] However, in the existing image recognition-based automatic inventory, traceability and deep learning labeling processes for instruments, different instrument materials react differently to complex ambient light and are difficult to accurately predict. For example, instruments made of stainless steel, titanium alloy, ceramics and other materials often experience coexistence of strong specular highlights and diffuse reflections under variable lighting conditions, which can easily lead to overexposure of local areas and obstruction of instrument surface texture information, resulting in missed detections in the target detection model, affecting recognition accuracy and stability, and making it difficult to meet high-standard clinical management needs. Summary of the Invention

[0004] The present invention provides a medical device intelligent identification method, device, equipment and medium to solve the technical problem of low accuracy in medical device intelligent identification.

[0005] In a first aspect, a method for intelligently identifying medical devices is provided, comprising: After receiving a medical device identification instruction from a user, controlling the image acquisition unit to acquire a first image of the medical device; Inputting the acquired first image into a predetermined first analysis model and determining it according to a preset confidence threshold to obtain a first confirmed image recognition result and a first uncertain image recognition result; For the local target area corresponding to the first uncertain image recognition result, controlling the image acquisition unit to obtain a second image of the medical device, and performing a de-glare adjustment process on the second image to obtain a glare-free image; Inputting the glare-free image into the first analysis model to obtain a second determined image recognition result of the medical device; A target medical device corresponding to the recognition instruction is determined according to the first determined image recognition result and the second determined image recognition result.

[0006] In a second aspect, a medical device intelligent identification device is provided, comprising: A first image acquisition module is configured to control the image acquisition unit to acquire a first image of the medical device after receiving a medical device identification instruction from a user; An image recognition module is configured to input the acquired first image into a predetermined first analysis model and determine the image according to a preset confidence threshold to obtain a first confirmed image recognition result and a first uncertain image recognition result; an image tuning and processing module, configured to control the image acquisition unit to acquire a second image of the medical device for the local target area corresponding to the first uncertain image recognition result, and perform a glare removal and tuning process on the second image to obtain a glare-free image; a second determined image recognition result analysis module, configured to input the glare-free image into the first analysis model to obtain a second determined image recognition result of the medical device; The medical device determination module is used to determine the target medical device corresponding to the recognition instruction according to the first determination image recognition result and the second determination image recognition result.

[0007] In a third aspect, a medical device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned medical device intelligent identification method when executing the computer program.

[0008] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned medical device intelligent identification method are implemented.

[0009] The present invention acquires comprehensive initial visual information by promptly capturing the first image after receiving the recognition instruction, and combines dual-polarization capture with pixel difference technology to accurately separate the specular highlight and diffuse reflection areas, thereby improving image clarity. The first analysis model is used to quickly determine the definite and uncertain recognition areas, and the second image of the uncertain area is focused on to reduce redundant acquisition, thereby improving data efficiency and accuracy. The image after de-glare adjustment processing has clearer feature information, and after re-analysis, it can significantly improve the recognition accuracy of the uncertain area. Finally, the first and second definite recognition results are integrated to comprehensively improve the accuracy and reliability of medical device recognition, thereby effectively solving the technical problem of insufficient accuracy of existing medical device intelligent recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] To facilitate the description of the technical solutions of the embodiments of the present invention, a brief description is now given of the drawings cited in the embodiments. It should be understood that the following drawings only show several typical embodiments of the present invention, and those skilled in the art can obtain other forms of drawings on this basis without creative work.

[0011] Figure 1 This is a schematic diagram of an application environment of a medical device intelligent identification method according to an embodiment of the present invention; Figure 2 This is a flow chart of a medical device intelligent identification method according to an embodiment of the present invention; Figure 3 is a schematic diagram of a dual-camera and polarization hardware assembly according to an embodiment of the present invention; Figure 4a 1 is a schematic diagram of a captured image with a polarization angle of 0° in one embodiment of the present invention; Figure 4b 1 is a schematic diagram of a captured image with a polarization angle of 90° in one embodiment of the present invention; Figure 4c 1 is a schematic diagram of a collected image with fusion and glare removal in one embodiment of the present invention; Figure 5a is a schematic diagram of dual polarization image acquisition in one embodiment of the present invention; Figure 5b is a schematic diagram of an image after the glare removal effect in one embodiment of the present invention; Figure 6 is a schematic diagram of the final verification set in one embodiment of the present invention; Figure 7 is a schematic diagram of a normalized confusion matrix in one embodiment of the present invention; Figure 8 is a schematic diagram of a medical device area marking according to an embodiment of the present invention; Figure 9a This is a schematic diagram showing that the device photo recognition and the difference counting and verification interface recognition are consistent in one embodiment of the present invention; Figure 9b Schematic diagram of the inconsistency between the device photo recognition and the count difference verification interface recognition in one embodiment of the present invention; Figure 10 This is a schematic structural diagram of a medical device intelligent identification device according to an embodiment of the present invention; Figure 11 is a structural schematic diagram of a medical device in one embodiment of the present invention; Figure 12 FIG. 1 is another structural diagram of a medical device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0012] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0013] The medical device intelligent identification method provided by the embodiment of the present invention can be applied in Figure 1 In an application environment, the client communicates with the server through a network. After receiving the medical device recognition instruction issued by the client, the server controls the image acquisition unit to obtain a first image of the medical device; the acquired first image is input into a predetermined first analysis model and judged according to a preset confidence threshold to obtain a first definite image recognition result and a first uncertain image recognition result; for the local target area corresponding to the first uncertain image recognition result, the image acquisition unit is controlled to obtain a second image of the medical device, and a de-glare adjustment process is performed on the second image to obtain a glare-free image; then, the glare-free image is input into the first analysis model to obtain a second definite image recognition result of the medical device; based on the first definite image recognition result and the second definite image recognition result, the target medical device corresponding to the medical device recognition instruction issued by the user is finally determined, and the recognition result is returned to the client. The present invention acquires comprehensive initial visual information by promptly acquiring the first image after receiving the recognition instruction, and combines dual-polarization acquisition with pixel difference technology to accurately separate the specular highlight and diffuse reflection area, thereby improving image clarity. The first analysis model is used to quickly determine the definite and uncertain recognition areas, focus on the second image of the uncertain area, reduce redundant acquisition, and improve data efficiency and accuracy. The image after de-glare adjustment processing has clearer feature information, and after re-analysis, the recognition accuracy of uncertain areas can be significantly improved. Finally, the first and second identification results are integrated to comprehensively improve the accuracy and reliability of medical device recognition, thereby effectively solving the technical problem of insufficient accuracy of existing medical device intelligent recognition. Among them, the client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented as an independent server or a server cluster composed of multiple servers. The present invention is described in detail below through specific embodiments.

[0014] See also Figure 2 As shown, Figure 2 A schematic flow chart of a medical device intelligent identification method provided in an embodiment of the present invention includes the following steps: S1. After receiving a medical device identification instruction from a user, control an image acquisition unit to acquire a first image of the medical device.

[0015] A typical application scenario for this invention is the current closed-loop management process for surgical instruments, where the instrument flow follows a chain of "dispatched from a sterilization and supply center - used in the operating room - and then returned to the sterilization and supply center." Each step relies heavily on manual verification of instrument type, quantity, and integrity, which carries risks of low efficiency, missed or incorrect clearances, and difficulty ensuring accurate instrument identification. Therefore, the present invention proposes to use an image acquisition unit in conjunction with an intelligent recognition algorithm to assist in instrument identification and verification. For example, during preoperative preparation in the supply room, the supply room nurse verifies the model, quantity, and integrity of each instrument against a paper "Surgical Instrument Inventory Sheet" before packaging and sterilization. During the turnover and distribution process, the sterilization packages are placed in a temporary sterile storage warehouse and then distributed to the corresponding operating room according to the surgical schedule. The operating room performs a three-step inventory, where the scrub nurse and the circulating nurse count and sign each instrument before surgery, before closing the surgery, and after surgery. For example, a tertiary surgery lasting 4–6 hours typically requires two nurses to complete the inventory in a total of 30–40 minutes. Any missing instrument may result in delayed closing or a second exploration. The supply room rechecks the instrument package after it returns to the contaminated area, where each instrument is checked again, noting any defects or items requiring repair before entering the next round of the cleaning, packaging, and sterilization process. All of these processes rely on manual verification, which is inefficient, carries the risk of missed or incorrect clearances, and makes it difficult to ensure accurate instrument identification. Therefore, it is necessary to implement intelligent instrument identification and assisted verification through medical devices.

[0016] In an embodiment of the present invention, dual cameras are set at a sterile safety height of 40-50 cm above the instrument (fixed by a sterilizable magnetic bracket or disposable plastic clamp to ensure that they do not touch the sterile area). Without the need to modify the operating table or inventory table, a conveniently deployed intelligent medical device identification solution can be provided for clinical use.

[0017] In specific implementation, users can actively issue medical device identification instructions in any of the following ways: Through the preset physical controls on the front-end image acquisition unit; Through the preset touch controls on the display interface of the front-end image acquisition unit; The backend server issues instructions through the human-computer interaction interface on the control terminal.

[0018] Upon receiving the recognition command, the image acquisition unit is controlled to capture a first image of the medical device. The image acquisition unit can be located in a medical device inventory area, a disinfection supply center inventory area, or an operating room medical device storage area. It is secured 40-50 cm above the sterile area using a sterilizable magnetic bracket or disposable clamp, allowing for easy deployment without disrupting the sterile environment. This first image is the valid medical device image exposed and output by the acquisition unit.

[0019] In an embodiment of the present invention, the image acquisition unit is controlled by the terminal and interacts with the server to complete the issuance of the shooting instruction. After receiving the medical device identification instruction issued by the user, the image acquisition unit is controlled to obtain the first image of the medical device, including: Determine the target recognition area based on user instructions; Sending shooting action instructions to the image acquisition unit in the area; The imaging unit is controlled according to the instruction to capture a first image of the medical device.

[0020] In detail, the medical device identification area refers to the physical space range that the image acquisition unit needs to focus on and shoot, which directly affects the integrity and identification efficiency of medical devices in the captured image, such as the medical device inventory area, the disinfection supply center inventory area or the operating room medical device placement area. In order to improve efficiency, when the identification instructions involve multiple areas and multiple devices, the system can simultaneously trigger multiple cameras to collect in parallel to reduce operation delays.

[0021] Specifically, a camera_trigger topic message is sent through the MQTT protocol. The message carries the area ID and shooting parameters, such as {"area_id":"OR-1-01", "mode":"high_res"}. The capture() interface of the camera SDK is called to directly trigger the image sensor to start frame acquisition, thereby obtaining the image corresponding to the medical device in the identification area.

[0022] Furthermore, the accuracy of medical device recognition is directly related to surgical safety. Missing recognition can lead to serious consequences, such as the device being left in the body. Therefore, medical scenarios require zero tolerance for recognition errors. This invention reduces the risk of missed and incorrect detections and improves recognition reliability through precise collection and identification.

[0023] S2. Input the acquired first image into a predetermined first analysis model and make a determination according to a preset confidence threshold to obtain a first determined image recognition result and a first uncertain image recognition result.

[0024] In an embodiment of the present invention, the first analysis model is a deep learning model trained on a large number of medical device images. It has a fixed network structure and weight parameters, typically using a lightweight YOLO-v8 network. It is usually obtained through supervised learning training, such as using a cross-entropy loss function. This model is used to extract and classify features from medical device images. Its backbone is based on a simplified CSPDarknet structure, which can extract image features such as edges and textures through a multi-layer convolutional network and fuse feature maps of different scales to meet the detection requirements of medical devices of different sizes. Ultimately, the model is normalized through a fully connected layer and a softmax layer, and outputs a probability distribution vector for each medical device category corresponding to the first image. The first determinate image recognition result refers to a high-confidence recognition result output by the model and can be used directly as a valid recognition result. The first uncertain image recognition result refers to a low-confidence result output by the model with ambiguous category distinctions. It cannot be used directly and requires manual review or a secondary recognition process.

[0025] In an embodiment of the present invention, inputting the acquired first image into a predetermined first analysis model and determining according to a preset confidence threshold to obtain a first determined image recognition result and a first uncertain image recognition result includes: Inputting the first image into the first analysis model, and outputting a probability distribution vector corresponding to the first image; When the highest probability value in the probability distribution vector is greater than or equal to the preset confidence threshold, determining the medical device category corresponding to the highest probability value as the target device category corresponding to the first image, and generating a first determined image recognition result according to the target device category; Calculate the difference between the highest probability value and the second highest probability value in the probability distribution vector; when the highest probability value is less than the preset confidence threshold and the difference is less than the preset difference threshold, determine the medical device category corresponding to the highest probability value as the candidate device category corresponding to the first image, and generate a first uncertain image recognition result based on the candidate device category.

[0026] In detail, the first analysis model refers to a deep learning model that has been pre-trained with massive medical device images. It usually uses a lightweight YOLO-v8 network, and its Backbone is based on a simplified CSPDarknet structure. The core is to efficiently extract image features, that is, it has the ability to extract efficient image features. The image tensor of the first image is reshaped and then the image edge and texture features are extracted through the convolution layer. That is, after the first image enters the network through the input layer, it first passes through multiple convolution blocks to gradually compress the spatial dimensions and increase the number of channels to capture low-level features such as edges and textures. The high-level semantic features output by the Backbone are passed downward and fused with the middle and low-level features to make up for the feature loss caused by the lightweight structure. Finally, three fused feature maps are generated, which are suitable for the detection needs of medical devices of different sizes. After the feature map is mapped by the fully connected layer, it is normalized by the softmax layer to output the probability distribution of the medical device category corresponding to the image.

[0027] Specifically, the preset confidence threshold and difference threshold are both system-preset and can be determined through experience or experimentation based on the distribution characteristics of the training data and the recognition accuracy requirements of actual medical scenarios. The preset confidence threshold is typically set between 0.6 and 0.9 to distinguish the reliability of the model output results. The confidence threshold can be dynamically adjusted based on the risk tolerance of different business scenarios and the distribution of the validation set data. For example, in scenarios with extremely high security requirements, such as "final inventory before closure of the cavity," the confidence threshold is typically raised to 0.85-0.90 to minimize the risk of missed detections. In scenarios with relatively low risks and high efficiency requirements, such as batch warehousing in disinfection supply centers, the threshold can be set between 0.60 and 0.70 to achieve faster turnover. The final threshold value is determined by combining true positive rate-false positive rate curve analysis, field test results, and business fault tolerance requirements. Results above this threshold can be directly used for inventory management or inventory records, while results below this threshold require manual review or secondary recognition. The difference threshold is used to assist in judging the discrimination of the model when the probabilities of multiple categories are close, avoiding forced judgment under low confidence, and improving security and accuracy.

[0028] Furthermore, if the highest probability output by the first analysis model is greater than or equal to a preset confidence threshold, and the categories are clearly distinguished, it is determined as a definite result, that is, the maximum value in the probability distribution and its corresponding category are queried, and whether the highest probability in the probability distribution vector is greater than or equal to the confidence threshold is verified. If satisfied, the corresponding category is determined as the target device category, and the image corresponding to the target device category is determined as the first definite image recognition result. The recognition result can be directly used as effective data for inventory management, inventory records and other business links, reducing manual intervention and improving work efficiency; if the highest probability value is lower than the threshold, and the difference with the second highest probability value is small, it indicates that the model has insufficient feature distinction ability for the image, which may be due to the similar appearance of the devices (such as different models of hemostatic forceps) or poor image quality (such as glare blocking texture). At this time, the preset probability difference threshold is used for screening to avoid inaccurate judgments on the image and reduce medical risks. For example, if the highest probability value in the probability distribution vector is greater than 0.8 and the category is obvious, it can be directly determined as the first confirmed image recognition result and used for automatic registration to improve inventory efficiency; if the highest probability value is lower than the preset confidence threshold and is close to the second highest probability value, it means that the model's ability to distinguish is insufficient (such as similar appearance of the devices or poor image quality), then it is determined as the first uncertain image recognition result, and it is necessary to enter the next step of the image tuning and recognition process to reduce medical risks. Furthermore, for the first uncertain image recognition result, subsequent image adjustment and re-analysis are required to obtain more accurate medical device recognition results to ensure the accuracy and reliability of the medical device recognition process and avoid omissions or misjudgments.

[0029] S3. For the local target area corresponding to the first uncertain image recognition result, control the image acquisition unit to obtain a second image of the medical device, and perform a de-glare adjustment process on the second image to obtain a glare-free image.

[0030] In an embodiment of the present invention, the local target area refers to the actual physical storage space of medical devices in a clinical environment, such as a fixed area (such as an instrument rack in a supply room) or a dynamic area (such as a movable instrument table in an operating room). The second image refers to an image collected again to supplement the identification information after the first uncertain result is generated, and includes at least two viewing angles with different polarization angles, or viewing angles at different physical positions. The main body of the image is consistent with the first image, but due to different polarization angles or positions, the brightness distribution of the mirror reflection area is different.

[0031] In an embodiment of the present invention, controlling the image acquisition unit to acquire a second image of the medical device for the local target area corresponding to the first uncertain image recognition result includes: Identifying image acquisition units at different positions in the local target area; Determining the shooting directions of image acquisition units at different positions according to the polarization angle of the preset rotation angle; The image acquisition units at different positions are controlled by the shooting direction to shoot images, thereby obtaining second images of the medical device in different shooting directions.

[0032] Specifically, multiple image acquisition units can be deployed in a localized target area, and two acquisition units can be controlled to capture images using polarizers at preset angles. For example, when the polarization angles of the two acquisition directions differ by 90°: θ1 = 0° (horizontal polarization) primarily captures the diffuse reflection component, while θ2 = 90° (vertical polarization) primarily captures the specular reflection component. Leveraging the physical properties of polarized light, images Iθ1 and Iθ2 are captured using 0° and 90° polarizers, respectively. Subsequently, the diffuse and specular reflection components can be separated through pixel-level differentiation.

[0033] Specifically, if Figure 3 As shown, two cameras with a resolution of ≥1920×1080 can be selected, with 0° / 90° CPL polarizers fixed in front of the lenses respectively. They are installed side by side 45 cm above the local target area. The cameras are connected to the host via a USB-3.0 Hub and automatic exposure and automatic white balance are turned off. When the first uncertain result is generated, the system starts the second image acquisition process, sets the polarization angles to θ1 and θ2 (such as 0° and 90°) to the corresponding acquisition units, and sends a synchronous trigger signal to ensure the consistency of the acquisition time, outputting Iθ1 and Iθ2.

[0034] In addition, if Figure 4a The image shown is captured when the polarization angle is 0°; Figure 4b The image shown is captured under a polarization angle of 90°. Figure 4c The figure shows the result of fusion and glare removal of two frames of images.

[0035] Furthermore, in order to ensure that the image tuning processing algorithm is effective for partial light reflection, enhance the adaptability to local reflection tuning of different materials, and while enhancing the processing of image detail features, the consumption of algorithm resources can be controlled to achieve an effective balance between recognition accuracy and computing resource consumption.

[0036] In the embodiment of the present invention, the glare-free image refers to a high-quality image that meets specific application requirements and is obtained after performing glare reduction optimization processing on the second image. In the embodiment of the present invention, performing the de-glare optimization process on the second image to obtain the glare-free image includes: identifying a first-viewing angle image and a second-viewing angle image corresponding to the polarization angle in the second image; Calculate the pixel-level difference between the two-view images and normalize them to obtain the mirror probability map; Calculating an adaptive light suppression weight based on the refractive index and specular reflectivity of the medical device material, and weighting the specular probability map to generate a de-glare image; Adjusting the color attributes of the de-glare image in a plurality of preset color spaces to obtain a target de-glare image corresponding to each color space; Calculating the comprehensive reflection intensity of the de-glare image and generating fusion weights corresponding to each color space in combination with the adaptive light suppression weights; Performing weighted fusion on the glare-removed images of each target according to the fusion weights to obtain an optimized processed image; Local refinement is performed on the high-mirror probability region of the tuned image to obtain a glare-free image.

[0037] Specifically, after the first uncertain image recognition result appears, the system collects two frames of to-be-fused computational images Iθ1 and Iθ2 at two preset polarization angles (e.g., 0° and 90°) in the same local target area. For example, the pixel-level difference between the two frames of to-be-fused computational images Iθ1 and Iθ2 can be calculated according to formula (1): ; Then, according to formula (2), the difference map corresponding to the pixel-level difference between Iθ1 and Iθ2 is Normalize: (2), where is the minimum value, is the specular probability map corresponding to Iθ1 and Iθ2, max D represents the maximum value of the difference map D in the entire image (i.e., max{i,j} D(i,j)), and the value range of w is [0,1]. After normalization, the specular probability map in the range of 0 to 1 is obtained (i.e., the pixel-level glare intensity mask), which is used for subsequent weight allocation and local refinement. In addition, in order to adapt to the reflective characteristics of instruments of different materials, the refractive index recorded in the instrument material library can be combined and the mirror reflectivity R s Calculate the adaptive weight α.

[0038] For example, it can be calculated according to formula (3): (3), where 、 is the weight coefficient obtained through experience or training, and the refractive index and the mirror reflectivity R s All of them are derived from the preset material parameter library: the library is based on the data of public optical manuals, the technical specifications of equipment manufacturers and a small amount of sampling calibration results, and can be continuously maintained by equipment category (stainless steel, titanium alloy, ceramic, etc.); if the corresponding material is temporarily missing in the library, the system will automatically use the default average value of the equipment in that category, and finally The normalized mapping is performed to the interval (0, 1) to obtain the adaptive weight α.

[0039] Furthermore, the mirror probability map (obtained by normalizing the difference graph D through formula (2)) and the adaptive weight Combined, the de-glare fusion is weightedly suppressed to generate the de-glare image F1, thereby achieving a unified de-glare effect among materials such as stainless steel, titanium alloy, and ceramics. Figure 5a As shown in the figure, the original image collected by dual polarization has obvious specular highlights; Figure 5b As shown, the de-glare image F1 obtained after the above-mentioned difference normalization and material adaptive weight processing basically eliminates glare.

[0040] Specifically, brightness compression and / or local equalization can be performed on the deglare image F1 in at least two preset color spaces (e.g., any two of RGB, HSV, LAB, and YUV) to obtain processing results for each color space. Subsequently, the highlight pixel ratio H_ratio,c and overexposed pixel ratio O_ratio,c of F1 after processing in each color space are calculated, and the reflective intensity of the color space is calculated: R_(intensity,c) = H_ratio,c + 2·O_ratio,c (4). The coefficient 2 is an empirical weight used to emphasize the impact of overexposure on glare. R_(intensity,c) is then compared with preset thresholds T1 and T2 to determine the reflective level (mild / medium / heavy) of the color space, and the fusion weight w_c of the color space is dynamically generated based on this.

[0041] For example, let the processing result of color space c (such as RGB) be F_c, and the processing result of color space d (such as HSV) be F_d: In highly reflective areas (pixel-level glare weight close to 1), F_d is preferred (HSV's V channel suppresses light more thoroughly), while in weakly reflective areas (pixel-level glare weight close to 0), F_c is preferred (RGB has high color fidelity). Based on this, the processing results of multiple color spaces are weighted fused to obtain the tuned processed image U. The reflective intensity R_(intensity,c) of the cth color space and the adaptive light suppression weight α_c of the corresponding material are used together in the weight calculation to obtain the fusion weight w_c of the cth color space, which is calculated as follows:

[0042] in Indicates the color space, To sum the subscripts, is the number of color spaces used; 、 is the weight coefficient, which is used to jointly adjust the influence of the reflection intensity and the light suppression adaptability of the material; R_(intensity,k) is the The reflection intensity of the color space, For the The adaptive light suppression weights for materials in each color space are applied. This multi-color space brightness compression and dynamic weight fusion mechanism automatically adjusts the output under different surgical light brightness and color temperature conditions, maintaining glare-free image brightness uniformity and reducing the need for recalibration.

[0043] Furthermore, multi-color space fusion has significantly reduced glare, but residual bright spots may still exist in strongly reflective areas on the surface of metal instruments (such as the tip of surgical forceps), affecting recognition accuracy. To this end, local optimization processing can be performed only on pixels whose mirror probability exceeds a preset threshold based on the mirror probability map.

[0044] In an embodiment of the present invention, in order to accurately identify an image in a highly reflective area on a metal instrument surface, performing local refinement on a high-mirror probability area of ​​the tuned image to obtain a glare-free image includes: Based on the specular probability map, a local filtering weight map is generated in combination with the adaptive light suppression weight; Extracting mirror pixels with a probability greater than a preset probability threshold from the mirror probability map as a high-probability pixel set; According to the local filtering weight map, edge-preserving smoothing is performed on the pixels in the high-probability pixel set to output a glare-free image.

[0045] In detail, the mirror probability map It is a grayscale image with the same size as the original image. Each pixel value represents the probability that the pixel belongs to the mirror reflection area. The probability can be obtained by pixel-level difference It is calculated that the larger the D value is, the higher the probability of the mirror is. Then the guided filter is used to perform edge-preserving smoothing on the mirror area: to tune the processed image U or its gradient map As a guide map G, the high mirror probability area is smoothed while retaining the instrument texture and edges to avoid overall blur.

[0046] Specifically, the mirror probability map is , let the probability threshold be T, when w(i,j)>T, the pixel is recorded as a high probability pixel (where i and j represent the pixel in the mirror probability map The coordinate index of the row and column direction is independent of the color channel), and T can be appropriately lowered for the edge area of ​​the device. Generate local filter weight map ,in , The larger the value, the stronger the specular reflection, and the smaller the corresponding weight, the greater the filtering intensity. In the area of ​​​​the filter, only weak filtering or no filtering is performed, and the weight map W is used to adjust the filtering strength and is combined with the guide map G (which can be U or ) work together to perform pixel-level correction on the tuned image and achieve directional optimization of the high mirror probability area. Furthermore, in an exemplary implementation, a Python + OpenCV CPU prototype implemented polarization differentiation and multi-color space processing at a resolution of 1280 × 720, with acceptable average processing time. Replacing the interpretation layer with C++ / SIMD optimization, retaining only two primary color spaces, and disabling debug logging can further reduce latency, meeting the requirements of real-time applications. After the aforementioned glare removal and color optimization, key image features are preserved or enhanced, necessitating the re-input of the glare-free image into the first analysis model for more accurate recognition results.

[0047] S4. Input the glare-free image into the first analysis model to obtain a second confirmed image recognition result of the medical device.

[0048] In the embodiment of the present invention, the second determined image recognition result is a recognition result obtained after analyzing the non-glare image. By optimizing the image quality, recognition ambiguity is reduced and the result is more reliable.

[0049] In detail, this step is consistent with the determination process described in S2: the results are still divided according to the preset confidence threshold and probability difference threshold, so it will not be described in detail.

[0050] Specifically, after taking the photo, the system inputs the glare-free image into the lightweight YOLO-v8 network, and instantly outputs the category, quantity, and preset confidence level of each device. For detection results with the highest probability value lower than the preset confidence threshold, the system will generate a review prompt on the user interface for the nurse to manually confirm as appropriate; the review records will be archived and can be used as reference data for subsequent model iterations and performance tuning.

[0051] For example, 1500 images of self-collected instruments (training: validation = 80:20) were selected and trained from scratch for 100 epochs on the YOLO-v8-s lightweight Backbone. The validation set results are as follows: Figure 6 As shown, mAP50 = 0.993, mAP50-95 = 0.933, Precision = 0.998, Recall = 0.995. The normalized confusion matrix is ​​as follows Figure 7As shown, the accuracy of Hemolok-L (clamping device - large) and Hemolok-XL (clamping device - extra large) is 100%, the accuracy of Hemolok-ML (clamping device - medium) is 97%, and the overall missed detection rate is 1.9%. The above data is only an example to illustrate the feasibility of this method and does not limit the scope of protection of the present invention.

[0052] Furthermore, the first image recognition result may be misclassified due to factors such as ambient light and occlusion, while the second image recognition result may be biased due to improperly set anti-glare processing parameters. To this end, the system can fuse and compare the two results: if the two are consistent, they are directly confirmed; if the category or quantity is inconsistent, the one with higher confidence is prioritized, or manual review is triggered to compensate for the error. For example, if the first result misclassifies the device category, the second result with higher confidence can be used to correct the error.

[0053] S5. Determine the target medical device corresponding to the recognition instruction according to the first determined image recognition result and the second determined image recognition result.

[0054] In the embodiment of the present invention, a more reliable recognition result is generated by the global coverage of the first determined image recognition result and the precise complementarity of the details of the second determined image recognition result.

[0055] In an embodiment of the present invention, determining the medical device corresponding to the medical device identification instruction issued by the user based on the first determined image recognition result and the second determined image recognition result includes: Performing result-level fusion on the first determined image recognition result and the second determined image recognition result to generate a fused recognition result; Parsing target device requirements from user identification instructions; The fusion recognition result is matched with the target requirement. When the category and quantity matching degree reaches the preset threshold, the target medical device corresponding to the recognition instruction is determined based on the fusion recognition result; otherwise, a difference prompt or manual review process is triggered.

[0056] In detail, the first determination image recognition result and the second determination image recognition result are weightedly superimposed at the result level according to the weights to generate a fusion recognition result, where the result includes the instrument category, quantity and location information to integrate the global information of the original image and the detail accuracy of the de-glare image, thereby reducing the probability of misjudgment caused by a single model; the recognition instruction input by the user may contain a text description (such as "scalpel") or a reference image, which must first be converted into a computable feature representation (feature vector), where the target instrument requirement is the target image / area specified in the instrument list or instruction.

[0057] Specifically, the similarity (e.g., cosine similarity) between the fusion result and the user command feature vector can be calculated. When this similarity is greater than or equal to a preset matching threshold, the two are considered consistent. If the threshold is not reached, processes such as reshooting or manual review can be triggered. This dual-result fusion mechanism can further reduce the risk caused by a single error.

[0058] Furthermore, in order to realize automatic omission correction, redundant correction, and placement correction functions, the method further includes: Marking the image area corresponding to the first image acquired by the control image acquisition unit in sequence to generate an actual recognition result and an actual marking sequence of the medical device; Identifying actual mapping relationship data between the actual recognition result and the actual tag sequence; generating target mapping relationship data based on target results and target tag sequences in a predetermined medical device inventory list; Analyzing whether there is a difference between the actual mapping relationship data and the target mapping relationship data; If there is a difference, a predetermined reminder unit is controlled to output reminder information corresponding to the difference.

[0059] Specifically, we first generate a target mapping sequence based on the selected instrument inventory, and mark all identifiable medical instruments (such as surgical forceps, syringes, gauze, etc.) in the first image. Specifically, we use the target detection algorithm to locate the bounding box of each instrument, and determine the marking order according to any of the following strategies: first, spatial coordinate sorting, with the upper left corner of the image as the origin, and numbering in the order of "left to right, top to bottom", which is suitable for fixed instrument table scenarios; second, functional usage order, numbering according to the order of use in the surgical process, which is suitable for dynamic operating room scenarios. For example, for an image containing three instruments, the actual marking sequence can be obtained: M001 - surgical forceps [50, 80, 150, 180], M002 - hemostatic forceps [200, 80, 300, 180], M003 - tweezers [350, 80, 450, 180] (see Figure 8 Based on the recognition output of the first analysis model, combined with the above-mentioned tag sequence, a one-to-one mapping relationship between the actual recognition result and the actual tag sequence can be formed, and the actual mapping relationship is used to record the "tag ID The corresponding information of "Device Category (and its Location)" is a quantitative description of the current device status. Correspondingly, the target mapping relationship is pre-set by the clinical specification or surgical process, including the sequential mapping of the target device category / quantity and its expected placement location.

[0060] Specifically, the system compares the actual mapping relationship with the target mapping relationship, comparing the quantity and sequence relationship between the two: ① Actual less than the target → omission; ② Actual more than the target → excess; ③ Category consistent but positional deviation → positional error, thereby determining whether there is a discrepancy. Specific examples include, but are not limited to, 1. Omission: The actual sequence lacks the device category or quantity specified in the target sequence; 2. Excess: The actual sequence contains device categories or quantities not specified in the target sequence; 3. Positional error: The category is consistent but the placement deviates significantly from the target. Positional determination can use metrics such as bounding box overlap and center point distance, and compare them with preset thresholds to determine whether there is a discrepancy. When a discrepancy is detected, the system controls the reminder unit to output a reminder message regarding the discrepancy. For example, the system may display four columns on the difference comparison result screen: "Device Name / Expected Quantity / Actual Detection / Status." The status column uses a √ / × icon to visually indicate consistency or discrepancy. After confirming, the nurse clicks "Return and Retake" or closes the reminder to complete the inventory cycle.

[0061] Furthermore, if Figure 9a As shown in the figure, the system will automatically pop up the recognition result interface after the photo is taken. The left side displays the viewfinder preview and operation controls such as "Identify Instrument". The right side displays the "No Difference" message in a green prompt bar and lists the difference details (instrument name, expected quantity, actual detection, status). The nurse can choose "Confirm" to end the current round of inventory, or "Retake Photo" to trigger the next recognition. The confirmation information can be written into the log for subsequent model iterations. Figure 9b As shown, when a discrepancy is detected, a red (or other prominent color) discrepancy bar will appear on the right side of the interface, and the specific abnormal item will be marked in the discrepancy table. The nurse can also choose to confirm or retake the photo, and the confirmation record can be archived for model optimization.

[0062] Furthermore, the system can generate different levels of alerts based on the discrepancy type: 1. Omission: A high-level warning is issued, and the target location is marked with a dotted box on the interface; 2. Redundancy: A general alert is provided, highlighting the bounding box of the redundant instrument; 3. Position Error: A prompt displays an overlay comparing the actual and target positions, with the offset noted. Alert content can be presented through text / image annotation on the display or by flashing the instrument table's LED indicator. Based on the medical risk level, omission corrections are prioritized, followed by redundancy corrections, and then position corrections.

[0063] Furthermore, in order to prevent errors in the second determined image recognition result and to prevent misidentification caused by extreme situations, the method further includes: generating an association tag report between the image and the recognition result based on the second determined image recognition result; Send the generated association mark report to the preset terminal and obtain the review results; The second determined image recognition result is determined or revised according to the audit result, and a final result is generated.

[0064] Specifically, the association tag report bridges the gap between machine recognition results and manual review. It presents recognition details in a visual manner, allowing reviewers to quickly determine correctness. The association tag report includes the image acquisition time, device ID, the recognition category of the second-determined image recognition result, the confidence level, and the positioning bounding box. For example, the association tag report is { "Report ID": "REP-20250627-001", "Recognition Result": { "Category": "Hemostat", "Confidence": 0.96, "Bounding Box": [210, 180, 320, 290] , "Risk Label": "Low Risk (No Obvious Occlusion)"}, "Related Data": {"Standard Template ID": "TPL-Hemostatic Forceps-003", "Feature Matching Degree": {"Overall": 0.91, "Forceps Head": 0.89, "Handle": 0.94}}, "Visualization Image Path": " / data / reports / vis_20250627-001.png"}.

[0065] Specifically, the report is sent to the audit terminal, and the interface can display a report visualization image on the left and comparison information with the standard template on the right. 1. Audit passed: The system records the auditor's ID and time, and marks the second confirmed image recognition result as final and valid. 2. Audit correction: The auditor can modify fields such as categories or bounding boxes, and the system updates the results accordingly and triggers model feedback. 3. Audit doubt: Marked as pending review and pushed to senior auditors for processing. The final results are written to the database and synchronized to downstream systems such as inventory records / inventory management.

[0066] It can be seen that the present invention significantly improves the accuracy and reliability of medical device identification by timely acquiring the first image, dual-polarization differential glare removal and secondary recognition fusion, and solves the problem of insufficient accuracy of the existing technology.

[0067] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0068] In one embodiment, a medical device intelligent identification device is provided, which corresponds to the medical device intelligent identification method in the above embodiment. Figure 10 As shown, the medical device intelligent identification device includes a first image acquisition module 101, an image recognition module 102, an image adjustment processing module 103, a second image recognition result analysis module 104 and a medical device identification module 105. The functional modules are described in detail as follows: The first image acquisition module 101 is configured to control the image acquisition unit to acquire a first image of the medical device after receiving a medical device identification instruction from a user; The image recognition module 102 is configured to input the acquired first image into a predetermined first analysis model and determine the image according to a preset confidence threshold to obtain a first confirmed image recognition result and a first uncertain image recognition result; The image optimization processing module 103 is configured to control the image acquisition unit to acquire a second image of the medical device for the local target area corresponding to the first uncertain image recognition result, and perform a glare reduction optimization process on the second image to obtain a glare-free image; A second determined image recognition result analysis module 104 is configured to input the glare-free image into the first analysis model to obtain a second determined image recognition result of the medical device; The medical device determination module 105 is configured to determine a target medical device corresponding to the recognition instruction according to the first determination image recognition result and the second determination image recognition result.

[0069] In one embodiment, the image recognition module 102, when inputting the acquired first image into a predetermined first analysis model and determining according to a preset confidence threshold to obtain a first determined image recognition result and a first uncertain image recognition result, is configured to: Inputting the first image into the first analysis model, and outputting a probability distribution vector corresponding to the first image; When the highest probability value in the probability distribution vector is greater than or equal to the preset confidence threshold, determining the medical device category corresponding to the highest probability value as the target device category corresponding to the first image, and generating a first determined image recognition result according to the target device category; Calculate the difference between the highest probability value and the second highest probability value in the probability distribution vector; when the highest probability value is less than the preset confidence threshold and the difference is less than the preset difference threshold, determine the medical device category corresponding to the highest probability value as the candidate device category corresponding to the first image, and generate a first uncertain image recognition result based on the candidate device category.

[0070] In one embodiment, the image optimization processing module 103, when controlling the image acquisition unit to acquire a second image of the medical device for the local target area corresponding to the first uncertain image recognition result, is configured to: Identifying image acquisition units at different positions in the local target area; Determining the shooting directions of image acquisition units at different positions according to the polarization angle of the preset rotation angle; The image acquisition units at different positions are controlled by the shooting direction to shoot images, thereby obtaining second images of the medical device in different shooting directions.

[0071] In one embodiment, the image optimization processing module 103, when performing the glare reduction optimization process on the second image to obtain the glare-free image, is configured to: identifying a first-viewing angle image and a second-viewing angle image corresponding to the polarization angle in the second image; Calculate the pixel-level difference between the two-view images and normalize them to obtain the mirror probability map; Calculating an adaptive light suppression weight based on the refractive index and specular reflectivity of the medical device material, and weighting the specular probability map to generate a de-glare image; Adjusting the color attributes of the de-glare image in a plurality of preset color spaces to obtain a target de-glare image corresponding to each color space; Calculating the comprehensive reflection intensity of the de-glare image and generating fusion weights corresponding to each color space in combination with the adaptive light suppression weights; Performing weighted fusion on the glare-removed images of each target according to the fusion weights to obtain an optimized processed image; Local refinement is performed on the high specular probability region of the tuned image to output a glare-free image.

[0072] In one embodiment, the image tuning module 103, when performing local refinement on the high-mirror-probability region of the tuned image and outputting the glare-free image, is further configured to: Based on the specular probability map, a local filtering weight map is generated in combination with the adaptive light suppression weight; Extracting mirror pixels with a probability greater than a preset probability threshold from the mirror probability map as a high-probability pixel set; According to the local filtering weight map, edge-preserving smoothing is performed on the pixels in the high-probability pixel set to output a glare-free image.

[0073] In one embodiment, when performing intelligent identification of medical devices, it is also used to: Marking the image area corresponding to the first image acquired by the control image acquisition unit in sequence to generate an actual recognition result and an actual marking sequence of the medical device; Identifying actual mapping relationship data between the actual recognition result and the actual tag sequence; generating target mapping relationship data based on target results and target tag sequences in a predetermined medical device inventory list; Analyzing whether there is a difference between the actual mapping relationship data and the target mapping relationship data; If there is a difference, a predetermined reminder unit is controlled to output reminder information corresponding to the difference.

[0074] In one embodiment, when performing intelligent identification of medical devices, it is also used to: generating an association tag report between the image and the recognition result based on the second determined image recognition result; Send the generated association mark report to the preset terminal and obtain the review results; The second determined image recognition result is determined or revised according to the audit result, and a final result is generated.

[0075] The present invention provides an intelligent identification device for medical devices. By promptly capturing a first image after receiving an identification instruction, comprehensive initial visual information is obtained, and dual-polarization acquisition and pixel difference technology are combined to accurately separate specular highlights and diffuse reflection areas, thereby improving image clarity. The first analysis model is used to quickly determine the definite and uncertain identification areas, and the second image of the uncertain area is focused on to reduce redundant acquisition, thereby improving data efficiency and accuracy. The image after de-glare adjustment processing has clearer feature information, and after re-analysis, the recognition accuracy of the uncertain area can be significantly improved. Finally, the first and second definite identification results are integrated to comprehensively improve the accuracy and reliability of medical device identification, thereby effectively solving the technical problem of insufficient accuracy of existing medical device intelligent identification.

[0076] For the specific definition of the medical device intelligent identification device, please refer to the definition of the medical device intelligent identification method above and will not be repeated here. The various modules in the above-mentioned medical device intelligent identification device can be implemented in whole or in part through software, hardware, or a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the medical device in hardware form, or can be stored in the memory of the medical device in software form, so that the processor can call and execute the corresponding operations of the above modules.

[0077] In one embodiment, a medical device is provided. The medical device may be a server, and its internal structure diagram may be as follows: Figure 11 As shown. The medical device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the medical device is used to provide computing and control capabilities. The memory of the medical device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the medical device is used to communicate with an external client via a network connection. When the computer program is executed by the processor, it realizes the functions or steps on the service side of a medical device intelligent identification method.

[0078] In one embodiment, a medical device is provided. The medical device may be a client, and its internal structure diagram may be as follows: Figure 12 As shown. The medical device includes a processor, memory, network interface, display screen and input device connected via a system bus. The processor of the medical device is used to provide computing and control capabilities. The memory of the medical device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the medical device is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps on the client side of a medical device intelligent identification method.

[0079] In one embodiment, a medical device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the following steps are performed: After receiving a medical device identification instruction from a user, controlling the image acquisition unit to acquire a first image of the medical device; Inputting the acquired first image into a predetermined first analysis model and determining it according to a preset confidence threshold to obtain a first confirmed image recognition result and a first uncertain image recognition result; For the local target area corresponding to the first uncertain image recognition result, controlling the image acquisition unit to obtain a second image of the medical device, and performing a de-glare adjustment process on the second image to obtain a glare-free image; Inputting the glare-free image into the first analysis model to obtain a second determined image recognition result of the medical device; A target medical device corresponding to the recognition instruction is determined according to the first determined image recognition result and the second determined image recognition result.

[0080] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: After receiving a medical device identification instruction from a user, controlling the image acquisition unit to acquire a first image of the medical device; Inputting the acquired first image into a predetermined first analysis model and determining it according to a preset confidence threshold to obtain a first confirmed image recognition result and a first uncertain image recognition result; For the local target area corresponding to the first uncertain image recognition result, controlling the image acquisition unit to obtain a second image of the medical device, and performing a de-glare adjustment process on the second image to obtain a glare-free image; Inputting the glare-free image into the first analysis model to obtain a second determined image recognition result of the medical device; A target medical device corresponding to the recognition instruction is determined according to the first determined image recognition result and the second determined image recognition result.

[0081] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or medical device can be found in the relevant descriptions of the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0082] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0083] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0084] It should be noted that if software tools or components other than those of our company appear in the embodiments of this application, they are only used for illustration and do not represent actual use.

[0085] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A medical device intelligent identification method, characterized in that: include: After receiving a medical device identification instruction from a user, controlling the image acquisition unit to acquire a first image of the medical device; Inputting the acquired first image into a predetermined first analysis model and determining it according to a preset confidence threshold to obtain a first confirmed image recognition result and a first uncertain image recognition result; For the local target area corresponding to the first uncertain image recognition result, controlling the image acquisition unit to obtain a second image of the medical device, and performing a de-glare adjustment process on the second image to obtain a glare-free image; Inputting the glare-free image into the first analysis model to obtain a second determined image recognition result of the medical device; A target medical device corresponding to the recognition instruction is determined according to the first determined image recognition result and the second determined image recognition result.

2. The medical device intelligent identification method according to claim 1, characterized in that: The step of inputting the acquired first image into a predetermined first analysis model and determining according to a preset confidence threshold to obtain a first confirmed image recognition result and a first uncertain image recognition result includes: Inputting the first image into the first analysis model, and outputting a probability distribution vector corresponding to the first image; When the highest probability value in the probability distribution vector is greater than or equal to the preset confidence threshold, determining the medical device category corresponding to the highest probability value as the target device category corresponding to the first image, and generating a first determined image recognition result according to the target device category; Calculate the difference between the highest probability value and the second highest probability value in the probability distribution vector; when the highest probability value is less than the preset confidence threshold and the difference is less than the preset difference threshold, determine the medical device category corresponding to the highest probability value as the candidate device category corresponding to the first image, and generate a first uncertain image recognition result based on the candidate device category.

3. The medical device intelligent identification method according to claim 1, characterized in that: The controlling the image acquisition unit to acquire a second image of the medical device for the local target area corresponding to the first uncertain image recognition result includes: Identifying image acquisition units at different positions in the local target area; Determining the shooting directions of image acquisition units at different positions according to the polarization angle of the preset rotation angle; The image acquisition units at different positions are controlled by the shooting direction to shoot images, thereby obtaining second images of the medical device in different shooting directions.

4. The medical device intelligent identification method according to claim 3, characterized in that: The performing the de-glare adjustment processing on the second image to obtain a glare-free image includes: identifying a first-viewing angle image and a second-viewing angle image corresponding to the polarization angle in the second image; Calculate the pixel-level difference between the two-view images and normalize them to obtain the mirror probability map; Calculating an adaptive light suppression weight based on the refractive index and specular reflectivity of the medical device material, and weighting the specular probability map to generate a de-glare image; Adjusting the color attributes of the de-glare image in a plurality of preset color spaces to obtain a target de-glare image corresponding to each color space; Calculating the comprehensive reflection intensity of the de-glare image and generating fusion weights corresponding to each color space in combination with the adaptive light suppression weights; Performing weighted fusion on the glare-removed images of each target according to the fusion weights to obtain an optimized processed image; Local refinement is performed on the high specular probability region of the tuned image to output a glare-free image.

5. The medical device intelligent identification method according to claim 4, characterized in that: The performing local refinement on the high specular probability region of the tuned image to output a glare-free image includes: Based on the specular probability map, a local filtering weight map is generated in combination with the adaptive light suppression weight; Extracting mirror pixels with a probability greater than a preset probability threshold from the mirror probability map as a high-probability pixel set; According to the local filtering weight map, edge-preserving smoothing is performed on the pixels in the high-probability pixel set to output a glare-free image.

6. The medical device intelligent identification method according to claim 1, characterized in that: The method further comprises: Marking the image area corresponding to the first image acquired by the control image acquisition unit in sequence to generate an actual recognition result and an actual marking sequence of the medical device; Identifying actual mapping relationship data between the actual recognition result and the actual tag sequence; generating target mapping relationship data based on target results and target tag sequences in a predetermined medical device inventory list; Analyzing whether there is a difference between the actual mapping relationship data and the target mapping relationship data; If there is a difference, a predetermined reminder unit is controlled to output reminder information corresponding to the difference.

7. The medical device intelligent identification method according to claim 1, characterized in that: The method further comprises: generating an association tag report between the image and the recognition result based on the second determined image recognition result; Send the generated association mark report to the preset terminal and obtain the review results; The second determined image recognition result is determined or revised according to the audit result, and a final result is generated.

8. A medical device intelligent identification device, characterized in that: include: A first image acquisition module is configured to control the image acquisition unit to acquire a first image of the medical device after receiving a medical device identification instruction from a user; An image recognition module is configured to input the acquired first image into a predetermined first analysis model and determine the image according to a preset confidence threshold to obtain a first confirmed image recognition result and a first uncertain image recognition result; an image tuning and processing module, configured to control the image acquisition unit to acquire a second image of the medical device for the local target area corresponding to the first uncertain image recognition result, and perform a glare removal tuning process on the second image to obtain a glare-free image; a second determined image recognition result analysis module, configured to input the glare-free image into the first analysis model to obtain a second determined image recognition result of the medical device; The medical device determination module is used to determine the target medical device corresponding to the recognition instruction according to the first determination image recognition result and the second determination image recognition result.

9. A medical device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, it is used to implement the medical device intelligent identification method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it is used to implement the medical device intelligent identification method according to any one of claims 1 to 7.

Images