Interaction system of delivery window and control method thereof

Abstract

The invention discloses an interactive system of a delivery window and a control method of the interactive system. The control method comprises the steps of judging whether identity information of a user is matched with pre-stored information or not; when the identity information is matched, the electrical control module controls the delivery window electromagnetic lock to be unlocked; after it is detected that the delivery window electromagnetic lock is closed, article image data are collected, an image recognition model obtained based on multi-layer convolutional neural network training is used for classifying the article image data, and article classification information is obtained; judging whether the article classification information is matched with pre-stored information or not; when the article classification information is matched, starting sterilization; and obtaining sterilization time according to the sterilization state information, and stopping sterilization when the sterilization time reaches preset sterilization time. Through identity verification, convolutional neural network-based article identification, instruction interaction and sterilization control technologies, the problems of high safety risk, low operation efficiency and poor traceability of a traditional delivery window system are solved, and intelligent management of the delivery window is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of clean room sterilization, and specifically refers to an interaction system of a transfer window and its control method. Background Art

[0002] As the core equipment for item transfer in a clean environment, the transfer window is widely used in fields with extremely high requirements for aseptic operation such as medical treatment, biological laboratories, and pharmaceuticals. Its core function is to ensure the airtightness of the clean area when items are transferred between the internal and external environments and prevent cross-contamination. Ordinary transfer windows only have simple functions such as opening for sterilization and stopping for item retrieval. In the operation process, there is generally no identity verification for users and no item inspection, resulting in the following significant defects:

[0003] Insufficient security: Generally, mechanical buttons are used to open the door, lacking identity authentication for users.

[0004] Item identification relies on manual work: The judgment of the type of sterilized items depends on visual inspection by users, posing management risks, and lacking digital records, making it impossible to trace responsibilities.

[0005] The sterilization process is uncontrollable: The sterilization time and status lack dynamic monitoring. It may lead to incomplete sterilization due to equipment failures or human interruptions, and abnormal events (such as accidental opening of the door lock, power failure) cannot be real-time feedback, posing serious biosafety risks.

[0006] Therefore, there is an urgent need for an interaction system and control method for a transfer window that integrates intelligent identity verification, high-precision item identification, dynamic sterilization control, and effective data traceability to solve the safety hazards and efficiency bottlenecks of traditional technologies and meet the requirements of modern clean environments for intelligence, digitization, and traceability. Summary of the Invention

[0007] The purpose of the present invention is to provide an interaction system and control method for a transfer window. By combining an identity recognition module, an electrical control module, an image acquisition module, an information interaction module, and a central control module, it solves the problems of high safety risks and poor traceability caused by weak identity verification, high dependence on manual supervision, and uncontrollable sterilization process. Through the integration of intelligent identity verification, item recognition based on convolutional neural networks, a dual-confirmation instruction protocol, and closed-loop sterilization control, it realizes the intelligent management of the transfer window and improves the safety and standardization of the clean laboratory.

[0008] To achieve this purpose, an interaction system of a transfer window designed by the present invention includes:

[0009] An information storage module is used to store the identity information of pre-stored transfer window users, pre-stored item type information, and preset sterilization time;

[0010] An identity recognition module is used to obtain the identity information of users;

[0011] The central control module is used to determine whether the identity information of the user matches the pre - stored identity information of the transfer window user;

[0012] The electrical control module, when the identity information matches, controls the opening of the electromagnetic lock of the transfer window. When an item is placed in the transfer window, the electromagnetic lock of the transfer window closes; the electrical control module is used to detect whether the electromagnetic lock of the transfer window is closed;

[0013] The image acquisition module is used to acquire the image data of the item after the electrical control module detects that the electromagnetic lock of the transfer window is closed;

[0014] The image recognition module classifies the item image data using an image recognition model trained based on a multi - layer convolutional neural network to obtain item classification information;

[0015] The central control module is used to determine whether the item classification information matches the pre - stored item type information. When the item information matches, the electrical control module starts sterilization, and the central control module obtains the sterilization status information;

[0016] The electrical control module is used to obtain the sterilization time according to the sterilization status information. When the sterilization time reaches the preset sterilization time, the sterilization stops;

[0017] The information interaction module is used to display information, including the current user identity information, item classification information, and sterilization status information.

[0018] The present invention also provides a control method for a transfer window, including the following steps:

[0019] Determine whether the identity information of the user matches the pre - stored identity information of the transfer window user;

[0020] When the identity information matches, the electrical control module controls the opening of the electromagnetic lock of the transfer window. When an item is placed in the transfer window, the electromagnetic lock of the transfer window closes;

[0021] After the electrical control module detects that the electromagnetic lock of the transfer window is closed, the image acquisition module acquires the item image data, and the image recognition module classifies the item image data using an image recognition model trained based on a multi - layer convolutional neural network to obtain item classification information;

[0022] Determine whether the item classification information matches the pre - stored item type information;

[0023] When the item information matches, the electrical control module starts sterilization, and the central control module obtains the current user identity information, item classification information, and sterilization status information;

[0024] The electrical control module obtains the sterilization time based on the sterilization status information, and stops sterilization when the sterilization time reaches the preset sterilization time.

[0025] The beneficial effects of the present invention are as follows:

[0026] 1) Through the closed-loop management of identity authentication, electromagnetic lock control, image recognition, and sterilization process, unauthorized operations or the introduction of dangerous goods are avoided, and the risk of cross-contamination is significantly reduced.

[0027] 2) Based on the convolutional neural network (CNN) algorithm, item recognition based on computer vision is realized, the workload of manual intervention is reduced, and the management risk is effectively reduced.

[0028] 3) Through the multi-level feedback of sterilization start response signals, sterilization process synchronization signals, and sterilization process error signals, the sterilization time and status are monitored in real time, and sterilization anomalies are intervened in real time to ensure the thoroughness of sterilization.

[0029] 4) Automatically record events such as failed identity matching, incorrect item classification, and sterilization anomalies, and support the auditing and traceability of operation logs.

[0030] 5) The sterilization status, item classification results, and alarm information are displayed in real time through the visualization interface, realizing the visualization of the transfer window status. Description of the Drawings

[0031] Figure 1 It is a schematic diagram of the interaction system architecture of a transfer window provided for an embodiment of the present invention;

[0032] Figure 2 It is a flowchart of a control method for a transfer window provided for an embodiment of the present invention;

[0033] Figure 3 It is a schematic diagram of the instruction interaction between the central control module and the electrical control module through a preset dual confirmation protocol provided for an embodiment of the present invention;

[0034] Figure 4 It is a schematic diagram of the sterilization feedback instruction sending process provided for an embodiment of the present invention;

[0035] Among them: 101 - identity recognition module; 102 - electrical control module; 103 - image acquisition module; 104 - information interaction module; 105 - central control module; 106 - information storage module; 107 - image recognition module. Detailed Embodiments

[0036] The following further describes the present invention in detail with reference to the drawings and specific embodiments:

[0037] Embodiment 1:

[0038] AsFigures 1 to 4 As shown in the figure, an interactive system for a transfer window, comprising:

[0039] An information storage module 106 is used to store the identity information of the pre-stored transfer window users, the pre-stored item type information, and the preset sterilization time;

[0040] An identity recognition module 101 is used to obtain the identity information of the user;

[0041] A central control module 105 is used to determine whether the identity information of the user matches the identity information of the pre-stored transfer window users;

[0042] When the identity information matches, an electrical control module 102 controls the opening of the electromagnetic lock of the transfer window. When an item is placed in the transfer window, the electromagnetic lock of the transfer window closes; the electrical control module 102 is used to detect whether the electromagnetic lock of the transfer window is closed;

[0043] An image acquisition module 103 is used to acquire item image data after the electrical control module 102 detects that the electromagnetic lock of the transfer window is closed;

[0044] An image recognition module 107 classifies the item image data by using an image recognition model obtained by training based on a multi-layer convolutional neural network to obtain item classification information;

[0045] The central control module 105 is used to determine whether the item classification information matches the pre-stored item type information. When the item information matches, the electrical control module 102 starts sterilization, and the central control module 105 obtains sterilization status information;

[0046] The electrical control module 102 is used to obtain the sterilization time according to the sterilization status information. When the sterilization time reaches the preset sterilization time, the sterilization is stopped.

[0047] In the interactive system of the transfer window in the present invention, through the closed-loop management of identity verification, electromagnetic lock control, image recognition, and sterilization process, the safety and standardization of the transfer window operation are significantly improved. Through identity matching, it is ensured that only authorized personnel can operate. The image classification technology uses a convolutional neural network (CNN) to achieve high-precision item recognition. Combining the physical control of the opening and closing of the electromagnetic lock prevents unauthorized items from entering and avoids dangerous or incorrect items from entering the sterilization link; the dynamic monitoring of the sterilization time and the preset threshold control ensure the thoroughness of sterilization. The process of the present invention has a high degree of automation, reduces the risk of human intervention, is applicable to high-clean environments such as medical and laboratory, and reduces the possibility of cross-contamination and operation errors.

[0048] In the above technical solution, the identity recognition module 101 includes an RFID card reading circuit.

[0049] In one embodiment, one component of the RFID card reading circuit is an RC522 non-contact reader / writer chip, which follows the ISO / IEC 14443 communication protocol, operates at 13.56 MHz, and reads the unique encoded UID stored in the nearby RFID tag.

[0050] In one embodiment, another component of the RFID card reading circuit is an STM32F103C8T6 main control chip, which communicates with the RC522 chip through the SPI communication protocol, returns the unique encoded UID stored in the RFID tag to the microcontroller through the SPI interface, and then sends it to the central control module 105 through the serial communication protocol.

[0051] The serial communication protocol can be RS232 or RS485. Preferably, it can be converted to the USB communication protocol using a CH340G or PL2303 chip.

[0052] In the above technical solution, the electrical control module 102 includes a microcontroller MCU, an opto-isolation circuit, a relay control circuit, and a serial communication circuit.

[0053] In one embodiment, the microcontroller MCU uses an STM32F103C8T6 main control chip, which is used to receive the instruction signal from the central control module 105, and operate the relay control circuit according to the preset control logic to achieve control actions such as opening and closing the electromagnetic lock of the transfer window and outputting the sterilization signal, and receive the feedback signal from the relay control circuit, such as the status information of the electromagnetic lock and the completion signal of the sterilization process.

[0054] In one embodiment, the core of the opto-isolation circuit is a TLP281 chip, which is used to isolate the signal path between the microcontroller MCU and the relay control circuit, and avoid the interference and fluctuations caused by high voltage and large current loads such as electromagnetic locks and sterilization equipment from affecting the microcontroller.

[0055] In one embodiment, the core of the relay control circuit is a ULN2003a chip, which receives the control signal from the microcontroller MCU after opto-isolation and converts it into an electrical signal capable of driving the relay to act, thereby controlling the suction and release of the relay. Specifically, the serial communication circuit uses MAX485 and SP232 to convert the TTL level signal of the microcontroller into RS-232 and RS-485 level signals for communication with the central control module 105.

[0056] In the above technical solution, the image acquisition module 103 includes an image sensor circuit, an image processing circuit, a USB interface circuit, a lens, and an optical system.

[0057] In one embodiment, the image sensor circuit and the image processing circuit are centered around the OV5640 chip, which supports outputting 5-megapixel photos and 1080p videos. It has a built-in image signal processor that can process the output raw image data, including noise reduction, color correction, and automatic exposure control.

[0058] In one embodiment, the USB interface circuit is centered around the CY7C68013A chip, which transmits the processed image data to the central control module 105 through the USB interface, following the USB2.0 protocol.

[0059] In one embodiment, the lens and the optical system are centered around a 2.8mm glass lens, with a maximum visual range of 150 degrees.

[0060] In the above technical solution, the system further includes an information interaction module 104 for presenting information, including:

[0061] Obtaining the information that the personnel identity information matching fails, and presenting that the identity information of the user does not match the pre-stored identity information of the transfer window user;

[0062] Obtaining the information that the item matching fails, and presenting that the item classification information does not match the pre-stored item type information;

[0063] Presenting the current user identity information, item classification information, and sterilization status information, where the sterilization status information includes whether sterilization is enabled and the countdown to the preset sterilization time.

[0064] The information interaction module 104 further includes a function of collecting information, which supports manual input of the item name and quantity, supports receiving data input from external devices, and can transfer the collected data to other modules.

[0065] In the above technical solution, the information interaction module 104 includes a computer with a touch screen and a remote terminal computer.

[0066] In one embodiment, the computer with a touch screen is installed in the transfer window in an embedded installation manner, and has RS-232, RS-485, USB, and RJ45 interfaces, and can communicate with RS-232, RS-485, USB, and TCP / IP communication protocols.

[0067] Specifically, the remote terminal computer is a desktop PC, which is connected through a network cable and can communicate with the HTTP protocol to achieve remote monitoring and management. In the centralized monitoring scenario, the management personnel can view the usage status of multiple transfer windows in real time on the remote terminal and obtain information in a timely manner.

[0068] In the above technical solution, the central control module 105 is a computer with a touch screen.

[0069] In one embodiment, the central control module 105 may share the same computer with a touch screen with the information interaction module 104, adopt the processor and instruction system of the computer, and the operating system adopts Windows or Linux.

[0070] In one embodiment, the image recognition module 107 adopts a deep learning algorithm.

[0071] In one embodiment, the deep learning algorithm adopts a multi-layer convolutional neural network (CNN) model, which is pre-trained and deployed in the computer with a touch screen of the central control module 105.

[0072] In the above technical solution, the information storage module 106 can be deployed in a local hard disk, a distributed database or a cloud storage system to ensure high availability and scalability of data.

[0073] In one embodiment, the information storage module 106 may share the same computer with a touch screen with the central control module 105, and use the non-volatile memory of the computer, such as a hard disk.

[0074] In the electrical control module 102, a transfer window door opening timer is also set. After the electrical control module 102 receives the second confirmation instruction to open the transfer window electromagnetic lock and opens the transfer window electromagnetic lock, the electrical control module 102 starts the timer. When the time of the timer reaches the preset transfer window electromagnetic lock opening time, an alarm signal is sent to the central control module 105.

[0075] The information storage module 106 further includes a preset transfer window electromagnetic lock opening time.

[0076] The information interaction module 104 can receive the alarm signal and display a prompt to close the door as soon as possible.

[0077] In the above technical solution, when the user identity information does not match the pre-stored transfer window user identity information, the central control module 105 records the information of failed personnel identity matching;

[0078] When the item classification information does not match the pre-stored item type information, the electrical control module 102 opens the transfer window electromagnetic lock, and the central control module 105 records the information of failed item matching.

[0079] By recording the information of failed identity and item matching, the traceability of the system is enhanced. Logs are recorded when the personnel identity does not match, which is convenient for subsequent investigation of illegal access or permission issues; when the item type does not match, the lock is automatically opened and recorded, avoiding the retention of wrong items. This mechanism strengthens the fault tolerance and compliance of the system, is applicable to scenarios that require strict supervision (such as biological laboratories), and ensures that all operation records are traceable.

[0080] In the above technical solution, when the identity information of the user matches the pre-stored identity information of the transfer window user, the central control module 105 sends an opening instruction for the electromagnetic lock of the transfer window to the electrical control module 102. After receiving the opening instruction for the electromagnetic lock of the transfer window, the electrical control module 102 controls the opening of the electromagnetic lock of the transfer window. The user places an item in the transfer window. After closing the window door of the transfer window, the electromagnetic lock of the transfer window closes.

[0081] When the electrical control module detects that the electromagnetic lock of the transfer window is closed, the electrical control module 102 sends a closing feedback instruction for the transfer window to the central control module 105. After receiving the closing feedback instruction for the transfer window, the central control module 105 sends an image acquisition instruction to the image acquisition module 103. After receiving the image acquisition instruction, the image acquisition module 103 acquires item image data. The central control module 105 sends the item image data to the image recognition module 107. The image recognition module 107 classifies the item image data using an image recognition model obtained by training based on a multi-layer convolutional neural network to obtain item classification information.

[0082] The central control module 105 matches the item classification information with the pre-stored item type information to determine whether the item classification information matches the pre-stored item type information:

[0083] If the item information matches, the central control module 105 sends a sterilization instruction to the electrical control module 102. After receiving the sterilization instruction, the electrical control module 102 outputs a sterilization signal, starts the sterilization equipment, and sends a sterilization feedback instruction to the central control module 105. After receiving the sterilization feedback instruction, the central control module 105 obtains the user identity information, item classification information, and sterilization status information. The electrical control module 102 obtains the sterilization time based on the sterilization status information. When the sterilization time reaches the preset sterilization time, it outputs a stop sterilization signal to stop the sterilization equipment; the user identity information, item classification information, and sterilization status information are used for information display.

[0084] If the item information does not match, the central control module 105 sends an opening instruction for the electromagnetic lock of the transfer window to the electrical control module 102 and records the item matching failure information.

[0085] The above solution realizes the efficient cooperation among various modules through electromagnetic lock control, sterilization start, and status feedback. As the core scheduling center, the central control module 105 ensures the strictness of the operation timing through the instruction chain. For example, the image is collected after the window is closed first, avoiding security vulnerabilities caused by incorrect step sequences. The closed-loop feedback (such as sterilization signal and status synchronization) and abnormal interruption mechanism (such as termination due to parsing failure) of the sterilization instruction improve the controllability of the sterilization process.

[0086] In the above technical solution, when the central control module 105 sends an instruction to open the electromagnetic lock of the transfer window, the central control module 105 and the electrical control module 102 perform instruction interaction through a preset dual confirmation protocol, including confirmation instructions and response instructions, as well as an instruction verification mechanism. The specific instruction interaction method is as follows:

[0087] The central control module 105 sends the first confirmation instruction to open the electromagnetic lock of the transfer window to the electrical control module 102. After receiving it, the electrical control module 102 sends a response instruction, and the central control module 105 uses the instruction verification mechanism to analyze the response instruction:

[0088] If the analysis fails or the response instruction is not received, the instruction interaction is aborted;

[0089] If the analysis is successful, the central control module 105 normally sends the second confirmation instruction to open the electromagnetic lock of the transfer window;

[0090] After receiving the second confirmation instruction to open the electromagnetic lock of the transfer window, the electrical control module 102 opens the electromagnetic lock of the transfer window.

[0091] Adopting a dual verification mechanism of "first confirmation - response - second confirmation" can effectively prevent the abnormal opening of the electromagnetic lock caused by misoperation or communication interference. Through the strict process of instruction verification and response analysis, the integrity and authenticity of instruction transmission are ensured. Avoid potential safety hazards caused by single instruction errors or losses. Combining the process abort design when verification fails further enhances the robustness of the system.

[0092] In the above technical solution, the sterilization feedback instruction includes a sterilization start response signal, a sterilization process synchronization signal, and a sterilization process error signal, specifically as follows:

[0093] The sterilization start response signal sent by the electrical control module 102 is analyzed by the central control module 105:

[0094] If the analysis of the sterilization start response signal fails, the transmission is terminated;

[0095] If the analysis of the sterilization start response signal is successful, a process synchronization signal is sent;

[0096] The electrical control module 102 periodically sends a sterilization process synchronization signal to the central control module 105 through a preset synchronization information sending period. The process synchronization signal uses a timestamp mechanism based on the system time; the system refers to the operating system or software system of the central control module.

[0097] When the electrical control module 102 executes the sterilization instruction, it detects whether there is an abnormality in the sterilization process by transmitting the feedback of the transfer window door state or the power-off time feedback. If an abnormality is detected during the sterilization process, the electrical control module 102 sends a sterilization process error signal to the central control module 105. After receiving the sterilization process error signal, the central control module 105 obtains the abnormal information and sends it to the information storage module 106 for storage.

[0098] The information interaction module 104 can receive the abnormal information and display the abnormality in the sterilization process.

[0099] The abnormalities include: forcibly opening the locked transfer window door during the sterilization process, and the transfer window door magnetic sensor outputs an error signal; or powering off before the sterilization is completed. After the electrical control module is powered on and restored, it reads the previous sterilization information and issues an error signal after judging it as abnormal.

[0100] Through hierarchical feedback signals (sterilization start response signal, sterilization process synchronization signal, and sterilization process error signal) and the timestamp mechanism, the full life cycle monitoring of the sterilization process is realized. The periodic synchronization signal combined with the system timestamp facilitates the central module to track the sterilization progress in real time and troubleshoot time deviation; the error signal accurately locates the root cause of the problem and quickly triggers an emergency response. The automatic recording and storage of abnormal information (such as insufficient duration, forced door opening) provide data support for post-event analysis and equipment maintenance, significantly reducing the risk of sterilization failure and improving the compliance of the sterilization process.

[0101] In one embodiment, the timestamp mechanism refers to the Unix timestamp, which refers to the total number of milliseconds from 00:00:00 on January 1, 1970, Greenwich Mean Time (08:00:00 on January 1, 1970, Beijing time) to the present. It is generally a 13-digit number, such as 1741910649214. This data can be converted to 2025-03-14 08:04:09.

[0102] In one embodiment, the electrical control module 102 and the central control module 105 are two different hardware. If the countdown of 15 minutes starts simultaneously on both sides, there may be a time error due to hardware performance, software delay, or the accumulation of errors (multiple sterilizations). The result is, for example, that the central control module 105 believes that the countdown has ended, but the electrical module is still working.

[0103] Therefore, each time the countdown is started, it is necessary to synchronize the time once to align the electrical control module 102 with the central control module 105 to a standard time node.

[0104] In the above technical solution, the formats of the sterilization instruction, the first confirmation instruction, the second confirmation instruction, and the transfer window closing feedback instruction are a fixed frame header, an instruction type, and a check bit; the formats of the response instruction and the response signal are a fixed frame header, an instruction type, a response result, and a check bit;

[0105] The format of the image acquisition instruction is a fixed frame header, an instruction type, acquisition parameters, and a check bit;

[0106] In the sterilization instruction, the fixed frame header is used to indicate the start of communication, the instruction type is used to indicate the start of sterilization, and the check bit is used to verify the sterilization instruction and ensure the integrity of the communication transmission;

[0107] In the first confirmation instruction, the fixed frame header is used for communication start, the instruction type is used to indicate a request to open the electromagnetic lock, and the check bit is used to verify the first confirmation instruction and ensure the integrity of the communication transmission;

[0108] In the response instruction, the fixed frame header is used to indicate the start of communication, the instruction type is used to indicate the response to the request to open the electromagnetic lock, the response result is used to indicate the successful reception of the request to open the electromagnetic lock, and the check bit is used to verify the response instruction and ensure the integrity of the communication transmission;

[0109] In the second confirmation instruction, the fixed frame header is used to indicate the start of communication, the instruction type is used to confirm the opening of the electromagnetic lock, and the check bit is used to verify the second confirmation instruction and ensure the integrity of the communication transmission;

[0110] In the transfer window closing feedback instruction, the fixed frame header is used to indicate the start of communication, the instruction type is used to indicate the window door closing feedback, and the check bit is used to verify the transfer window closing feedback instruction and ensure the integrity of the communication transmission;

[0111] In the image acquisition instruction, the fixed frame header is used to indicate the start of communication, the instruction type is used to indicate an image acquisition request, the acquisition parameters are used to set the resolution and frame rate, and the check bit is used for the image acquisition instruction and ensures the integrity of the communication transmission;

[0112] In the response signal, the fixed frame header is used to indicate the start of communication, the instruction type is the response to the sterilization instruction, the response result is used to indicate the successful reception of the sterilization instruction, and the check bit is used for the response signal and ensures the integrity of the communication transmission;

[0113] The instruction verification mechanism adopts a cyclic redundancy check algorithm with an 8-bit check code.

[0114] In one embodiment, the instruction verification mechanism can also adopt a 16-bit cyclic redundancy check algorithm or parity check or LRC check.

[0115] The abbreviation of the cyclic redundancy check algorithm with an 8-bit check code is CRC-8.

[0116] By unifying the instruction format and verification mechanism, the standardization of the cross-module communication protocol is achieved. The fixed frame header and CRC-8 verification ensure the uniqueness of instruction recognition and data integrity; the clear division of the instruction type field simplifies the module parsing logic; the design of embedding parameters (resolution, frame rate) in the image acquisition instruction supports dynamic configuration and adapts to different scenario requirements. This standardized protocol reduces the complexity of system development, enhances module compatibility, and facilitates subsequent function expansion (such as adding new instruction types without reconstructing the communication framework).

[0117] In the above technical solution, the method for obtaining the item classification information is as follows:

[0118] Perform preprocessing on the item image data in the transfer window by first resizing the image and then normalizing it to obtain the preprocessed item image data;

[0119] The specific method for resizing the item image is: use the bilinear interpolation algorithm to resize all the item image data in the transfer window to the same pixel size;

[0120] The formula for the normalization operation is:

[0121]

[0122] where, x normalized represents the normalized x, x is the training image of the item image, mean is the brightness mean of the training image dataset of the item image, std is the brightness standard deviation of the training image dataset of the item image; the training image dataset of the item image refers to the labeled historical medical supply image dataset;

[0123] Input the preprocessed item image data into an image recognition model trained using a multi-layer convolutional neural network. The preprocessed item image data first undergoes local moving scanning through the convolutional layer in the image recognition model, and then the result of the convolutional operation is used as the output of the neural node in the corresponding area. Then, the output is made non-linear through the activation function, and the result output by the convolutional layer is further downsampled through the pooling layer to reduce the number of model parameters. Then, it passes through the fully connected layer, and finally, a feature vector of the bounding box position and its confidence is obtained.

[0124] The formula for the convolutional layer can be expressed as follows:

[0125] z(u, v) = ∑ i ∑ j x i,j ×k u-i,v-j + b;

[0126] where, x is the preprocessed item image, regarded as an m×m matrix, x i,j$x_{ij}$ is the pixel value of the preprocessed item image at position $(i, j)$, $(u, v)$ represents the current calculated position coordinates on the output feature map, $z(u, v)$ represents the convolution result obtained at position $(u, v)$ after applying the convolution kernel to the preprocessed item image, $k$ is a convolution kernel of size $n$, which is an $n\times n$ matrix, $b$ is the convolution layer bias term, and it slides in an $m\times m$ matrix with a stride of $s$. After each slide, it multiplies with the corresponding elements in the matrix, and $p$ circles of 0 values are padded on the outermost layer of the $m\times m$ matrix. Finally, a matrix of size $w$ can be obtained, which is the feature map output by the convolution layer. The calculation formula of $w$ is as follows:

[0127]

[0128] Among them, $m$ is the rank of the $x$ matrix, $p$ is the zero-padding value, and $s$ is the stride of the convolution kernel; after adding an overall bias term $b$ to the $w$ matrix, the output $z$ of the final convolution layer is obtained. After passing through multiple convolution layers as described above, feature extraction can be completed.

[0129] The pooling layer generally uses average pooling or max pooling to retain the most significant values in the features, which can effectively reduce the number of parameters and the feature dimension. After the pooling layer divides the feature map into blocks, it performs downsampling operations separately. Taking max pooling as an example, its formula is as follows:

[0130]

[0131] In the formula, $M(i', j')$ is the $i', j'$th element of the output result, max represents the operation of taking the maximum value, $z(i\times s + u, j\times s + v)$ is the element of the local window corresponding to the input and output $M(i', j')$ of the pooling layer, ensuring that every element is traversed, $s$ is the pooling window stride, $f$ is the pooling window size, and $\chi$, $\delta$ are the row and column offsets within the pooling window, which are used to traverse all positions within the pooling window. Through the above operations, the feature map passes through the pooling layer to obtain the pooled feature matrix.

[0132] The fully connected layer is the link that aggregates the finally pooled feature matrix and maps it to the required dimension. In the image recognition model trained based on a multi-layer convolutional neural network, it is mainly used to output the four coordinates of the item category, confidence, and the bounding box of the item image data into a feature vector. Its essence is to establish a mapping model between feature information and class labels. Its formula is as follows:

[0133] $y = Wx' + b_1$;

[0134] Wherein, x′ is the feature matrix after pooling, y is the output vector, that is, the bounding box, item category, and confidence of the item image data. The bounding box is represented by pixel coordinates [x1, y1, x2, y2], where x1 and y1 represent the upper left coordinates of the item bounding box, and x2 and y2 represent the lower right coordinates of the item bounding box; the confidence ranges from 0 to 1, indicating the credibility of the recognition result; W is the weight matrix, whose dimension is m′×n′, m′ is the number of channels, n′ is the dimension of the input vector x′, and b1 is the bias term of the fully connected layer.

[0135] The weight matrix W, the bias term b of the convolutional layer, and the bias term b1 of the fully connected layer together constitute the pre-trained parameters.

[0136] The bounding box, item category, and confidence of the item image data in the output vector y are used as item classification information, and the output vector y is saved as image data with annotation boxes and sent to the information interaction module 104.

[0137] The information interaction module 104 can obtain and display the image data with annotation boxes.

[0138] In one embodiment, the activation function can adopt the Relu or Sigmoid function.

[0139] In one embodiment, there can be multiple convolutional layers and pooling layers as described above, and multiple convolutional layers, pooling layers, and fully connected layers can be organized in a certain order and structure to form a multi-layer convolutional neural network structure, such as commonly used algorithms like Yolo or Faster-RCNN.

[0140] By adjusting the image size through bilinear interpolation and normalizing operation, the recognition error problem caused by shooting angle and lighting difference is solved. The normalization parameters are based on the statistical values of the dataset, making the distribution of the input data consistent with the training set and improving the generalization ability of the model; the output of the bounding box coordinates and confidence not only identifies the item category but also locates the item position, providing multi-dimensional data for subsequent operations (such as abnormal item location). This method significantly improves the image recognition accuracy (especially for small-sized or low-contrast items), and the preprocessing steps are lightweight, suitable for real-time processing on embedded devices.

[0141] In the above technical solution, the method of using the image recognition model trained based on the multi-layer convolutional neural network is as follows:

[0142] Preprocess the historical medical supply image dataset with annotations. The preprocessing method is: first adjust the image size, and then perform the normalization operation.

[0143] Input the preprocessed historical medical supply image dataset into the multi-layer convolutional neural network model for training. Generally, the weight parameters can be updated by the backpropagation method, and the process is as follows:

[0144] Before starting the training, first divide the dataset into a training set and a validation set according to a certain ratio (usually 8:2). Use the data in the training set to train the model and use the trained model to validate on the validation set.

[0145] At the same time, a loss function L needs to be defined to measure the difference between the model output result and the true value. Taking the mean squared error as an example, the formula is as follows:

[0146]

[0147] In the formula, y′ is the true value, is the model prediction value.

[0148] At the beginning of training, all the weight parameters in the model are randomly initialized, including the bias term b of the convolutional layer, the weight matrix W of the fully connected layer, etc. Then, calculate the output once using the data in the validation set According to the gradient change of and the custom learning rate η to update the weight parameters of the model. The gradient calculation formula is:

[0149]

[0150] Among them, L(θ) represents the loss function of the multi-layer convolutional neural network model under θ, represents the vector of the gradient of the loss function L(θ) with respect to all weight parameters θ, and the direction is the direction in which the loss function L(θ) grows fastest at θ. θ represents all weight parameters, expressed as θ = (θ1, θ2, …, θ n ), represents the partial derivatives of the function L(θ) with respect to the parameters θ1, θ2, …, θ n .

[0151] Use the gradient descent method to update all weight parameters. The parameter update formula is:

[0152]

[0153] Among them, η is the learning rate, θ t+1 and θ t respectively represent all the weight parameters of the multi-layer convolutional neural network model at the (t + 1)-th and t-th steps of iteration; L(θ t ) represents the loss function of the multi-layer convolutional neural network model under θ t , represents the vector of the gradient of the loss function L(θ t ) with respect to all weight parameters θ;

[0154] After each update, calculate the loss function value of the validation set. When the change amplitude of the loss function value of the validation set is less than the set threshold or the accuracy of the validation set no longer improves, the training ends. Save the network structure of the multi-layer convolutional neural network and a set of weight parameters with the best performance on the validation set as the image recognition model and pre-training parameters obtained by training based on the multi-layer convolutional neural network.

[0155] The image recognition model obtained by training based on the multi-layer convolutional neural network includes multiple alternating convolutional layers and pooling layers, and a fully connected layer at the end. The image recognition model obtained by training based on the multi-layer convolutional neural network also includes the size n of the convolution kernel k, the stride s, and the zero-padding value p in the convolutional layer, the activation function selected by the convolutional layer, the pooling method adopted by the pooling layer, and the input and output vector dimensions of the convolutional layer and the fully connected layer.

[0156] The size n of the convolution kernel k, the stride s, and the zero-padding value p in the convolutional layer are custom parameters.

[0157] The pre-training parameters include the weight matrix W of the fully connected layer, the bias term b1, and the bias terms b of all convolutional layers.

[0158] The above process is a method for an image recognition model obtained by training based on a multi-layer convolutional neural network. By using the combination of training set and validation set division and backpropagation to optimize the weight parameters, it ensures that the model fully learns the characteristics of medical items and avoids overfitting, improving the classification accuracy. This method supports the item recognition requirements of the image processing module and ensures the correct recognition of items placed in the transfer window.

[0159] Example 2:

[0160] An interaction system for a transfer window includes the following steps:

[0161] Judge whether the identity information of the user matches the pre-stored identity information of the transfer window user.

[0162] When the identity information matches, the electrical control module controls the opening of the electromagnetic lock of the transfer window. When an item is placed in the transfer window, the electromagnetic lock of the transfer window closes.

[0163] After the electrical control module detects that the electromagnetic lock of the transfer window is closed, the image acquisition module acquires the item image data, and the image recognition module classifies the item image data using the image recognition model obtained by training based on the multi-layer convolutional neural network to obtain the item classification information.

[0164] Judge whether the item classification information matches the pre-stored item type information.

[0165] When the item information matches, the electrical control module starts sterilization, and the central control module obtains the sterilization status information.

[0166] The electrical control module obtains the sterilization time based on the sterilization status information, and stops sterilization when the sterilization time reaches the preset sterilization time.

[0167] One embodiment of the method of the present invention includes the following steps:

[0168] Step 1: Obtain the identity information of the user. The current user identity information is sent to the central control module 105 through the identity recognition module 101.

[0169] In the scenario of a microbiology laboratory in a university, researchers need to frequently transfer experimental samples and equipment, and there are strict requirements for personnel permissions in different areas of the laboratory. The user wears a campus card containing an RFID tag. When approaching the transfer window, the identity recognition module 101 quickly and accurately obtains the identity information. The system determines whether the person has the permission to use the current transfer window for item transfer according to the pre-set permission rules.

[0170] In one embodiment, the user holds the identity recognition module 101 close to the transfer window. The RC522 non-contact reader / writer chip of the RFID card reading circuit senses and reads the nearby RFID tag signal at a frequency of 13.56 MHz according to the ISO / IEC14443 communication protocol. When the signal is successfully read, the unique code UID stored in the tag is transmitted to the connected STM32F 103C8T6 main control chip. The UID data is received through the SPI communication protocol and, after being processed by the internal program, is parsed into identity information in the form of 10-bit or 8-bit decimal numbers. Subsequently, using the CH340G or PL2303 chip, the identity information is converted from TTL level to the level signal supported by the USB communication protocol and sent to the central control module 105 through the USB interface.

[0171] Step 2: After receiving the user identity information, the central control module 105 obtains the pre-stored user identity information of the transfer window from the information storage module 106 and makes a comparison. The pre-stored user identity information of the transfer window stored in the information storage module 106 exists in the form of a database table. Each record contains fields such as the person's name and identity code. The central control module 105 compares the received identity code with the identity codes of all records in the database one by one.

[0172] In one embodiment, after the central control module 105 receives the identity information sent by the identity recognition module 101 in the form of 10 - bit or 8 - bit decimal digits through the USB interface, it first performs a preliminary format check on this information, checking whether the length of the received string is 10 bits or 8 bits, and whether each character in the string is a digit. If the format does not meet the requirements, this situation will be recorded as an invalid data reception event, and at the same time, a corresponding error prompt message will be sent to the information interaction module 104 to inform the user that there may be a problem with the identity information reading and the card needs to be swiped again. After confirming that the identity information format is correct, the central control module 105 calls the information in the information storage module 106 for identity information comparison.

[0173] In one embodiment, the database of the information storage module 106 uses MySQL, and the central control module 105 calls the SQL query interface to find the record matching the received identity code from the MySQL database table.

[0174] If a matching identity code is found, the central control module 105 sends a transfer window electromagnetic lock opening instruction to the electrical control module 102. The transfer window electromagnetic lock opening instruction includes two confirmation instructions and one response instruction.

[0175] In one embodiment, the format of the first confirmation instruction is: fixed frame header (2 bytes), instruction type (1 byte, the value of 0x01 indicates a request to open the electromagnetic lock), check bit (1 byte, generated using the CRC - 8 check algorithm). The electrical control module 102 parses the instruction. If the check bit is correct and the instruction type is recognized as a request to open the electromagnetic lock, it returns a response instruction. The format of the response instruction is: fixed frame header (2 bytes), instruction type (1 byte, the value of 0x02 indicates a response to the request to open the electromagnetic lock), response result (1 byte, the value of 0x00 indicates successful reception of the request), check bit (1 byte, generated using the CRC - 8 check algorithm). After receiving the response instruction, the central control module 105 parses the check bit and the response result. If both are correct, it sends the second confirmation instruction. The format of the instruction is: fixed frame header (2 bytes), instruction type (1 byte, the value of 0x03 indicates confirmation to open the electromagnetic lock), check bit (1 byte, generated using the CRC - 8 check algorithm). After receiving the second confirmation instruction and passing the check, the electrical control module 102 opens the transfer window electromagnetic lock. CRC - 8 is the abbreviation of the cyclic redundancy check algorithm for 8 - bit check codes.

[0176] If no matching identity code is found, the central control module 105 records the information indicating the failure of personnel identity information matching, including the time, reason for failure (identity authentication failed), etc. At the same time, it prompts the information interaction module 104. After the computer with a touch screen receives the information, a prompt box pops up on the screen to inform the user to operate again. The system returns to step one and waits for new identity information to be input.

[0177] Step 3: After waiting for an item to be placed and the window door to be closed, the electrical control module 102 sends a feedback instruction to the central control module 105. The central control module 105 sends an instruction to the image acquisition module 103. The image acquisition module 103 acquires an image and sends it to the image recognition module 107. The image recognition module 107 recognizes the placed item, generates item information, and then feeds it back to the central control module 105.

[0178] In one embodiment, the user places an item in the transfer window and closes the window door. The electrical control module 102 detects the change in the closed state of the window door through the input pin connected to the window door sensor, and then sends a feedback instruction to the central control module 105.

[0179] In one embodiment, the format of the feedback instruction is: fixed frame header (2 bytes), instruction type (1 byte, the value of 0x05 indicates the feedback of window door closure), check bit (1 byte, generated using the CRC-8 check algorithm). After the central control module 105 receives the feedback instruction and parses it correctly, it sends an image acquisition instruction to the image acquisition module 103. The format of the instruction is: fixed frame header (2 bytes), instruction type (1 byte, the value of 0x06 indicates the image acquisition request), acquisition parameters (2 bytes, for example, setting the image resolution to 0x0200 means 1080p, and the frame rate to 0x001E means 30fps), check bit (1 byte, generated using the CRC-8 check algorithm).

[0180] After the image acquisition module 103 receives the image acquisition instruction, it starts to acquire the image of the item in the transfer window. The processed image data is transmitted to the central control module 105 through the USB interface circuit.

[0181] In one embodiment, the acquired original image data (1280 * 720 pixels) is processed by the built-in image signal processor for noise elimination (using a 3×3 Gaussian filtering algorithm), color correction (according to the preset color correction matrix), automatic exposure control (adjusting the exposure parameters according to the image brightness histogram), etc.

[0182] In one embodiment, after the central control module 105 receives the image data, it forwards it to the image recognition module 107. The image recognition module 107 processes the image using an image recognition model obtained by training based on a multi-layer convolutional neural network (CNN).

[0183] In one embodiment, the dataset used for training the multi-layer convolutional neural network (CNN) model is provided by the Experimental Animal Center of Huazhong University of Science and Technology, and includes 1,200 medical supply images in a total of 20 categories. Each image is manually annotated by 3 professional experimentalists according to the standard of "Information Technology Artificial Intelligence Data Annotation Procedures for Machine Learning" T / CESA 1040-2019. The annotation content includes the item category and bounding box coordinates, and undergoes three rounds of cross-validation. During the model training process, data augmentation operations such as randomly rotating the original image (±15°), adjusting the brightness (±20%), and adding Gaussian noise (σ = 0.05) are performed.

[0184] In one embodiment, the images input into the multi-layer convolutional neural network (CNN) model need to be preprocessed, including: resizing the images to 224×224 pixels (using the bilinear interpolation algorithm) and normalizing operations;

[0185] The preprocessed image data is input into the multi-layer convolutional neural network (CNN) model. Feature extraction is performed through multiple convolutional layers (using 3×3 convolutional kernels with a stride of 1) by applying pre-trained parameters, and model inference is carried out through a pooling layer (using 2×2 max pooling) and a fully connected layer. During the inference process, the model classifies and identifies the image features according to the parameters obtained from training, and obtains the bounding box, item category, and confidence of the item image data;

[0186] Among them, the bounding box is represented by pixel coordinates [x1, y1, x2, y2], where x1 and y1 represent the upper left coordinates of the item bounding box, and x2 and y2 represent the lower right coordinates of the item bounding box; the item category corresponds to the category label of the item, such as "pipette", "drug packaging", "glasses", etc.; the confidence value ranges from 0 to 1, and 0.95, for example, represents the credibility of the recognition result.

[0187] The bounding box, item category, and confidence of the item image data are used as item classification information, and the bounding box, item category, and confidence of the item image data are saved as image data with annotation boxes and fed back to the central control module 105.

[0188] Step 4: The central control module 105 matches the item classification information with the pre-stored item type information in the information storage module 106. After successful matching, the central control module 105 sends a sterilization instruction to the electrical control module 102. After receiving the sterilization instruction, the electrical control module 102 outputs a sterilization signal to start sterilization and sends a sterilization feedback instruction to the central control module 105. At the same time, the central control module 105 sends the collected user identity information, item classification information, and sterilization status information to the information interaction module 104 for visual display. If the matching fails, the central control module 105 sends an instruction to open the transfer window electromagnetic lock to the electrical control module 102. After receiving the instruction to open the transfer window electromagnetic lock, the electrical control module 102 opens the transfer window electromagnetic lock. The central control module 105 records the information indicating the failure of user identity information matching and sends an alarm signal to the information interaction module 104 to remind the user to remove the corresponding item and then return to execute Step 3.

[0189] In one embodiment, after receiving the item classification information, the central control module 105 compares it with the pre-stored item type information in the information storage module 106. The pre-stored item type information in the information storage module 106 is also stored in the form of a database table, and each record contains fields such as item name, category, specification, and allow transfer flag. The central control module 105 compares the item category obtained by image recognition with each record in the database one by one.

[0190] If a matching item category is found and the item is allowed to be transferred (the value of the allow transfer flag field in the database is "yes"), the central control module 105 sends a sterilization instruction to the electrical control module 102.

[0191] In one embodiment, the format of the sterilization instruction is: fixed frame header (2 bytes), instruction type (1 byte, with a value of 0x07 indicating start sterilization), check bit (1 byte, generated using the CRC-8 check algorithm). After receiving the sterilization instruction and verifying it without error, the electrical control module 102 issues a sterilization start signal to start the sterilization equipment and returns a sterilization feedback instruction to the central control module 105. The sterilization feedback instruction includes a sterilization start response signal, a sterilization process synchronization signal, and a sterilization process error signal.

[0192] In one embodiment, the format of the sterilization start response signal is: fixed frame header (2 bytes), instruction type (1 byte, with a value of 0x08 indicating the response to the sterilization instruction), response result (1 byte, with a value of 0x00 indicating successful receipt of the sterilization instruction), check bit (1 byte, generated using the CRC-8 check algorithm).

[0193] After starting sterilization, the electrical control module 102 sends a sterilization process synchronization signal to the central control module 105 at a set time interval (such as every 5 seconds).

[0194] In one embodiment, the sterilization process synchronization signal adopts a timestamp mechanism based on the system time (the timestamp is accurate to milliseconds, for example, the timestamp value of 1632547890567 represents 15:31:30.567 milliseconds on September 25, 2021), ensuring time synchronization. The central control module 105 sends the collected user identity information, item classification information, and sterilization status information (the sterilization status information includes: whether sterilization is started, the countdown of the sterilization time, reading the time of the current central control module as the start time, in the format of "YYYY-MM-DD HH:MM:SS", where Y = year, M = month, D = day, H = hour, M = minute, S = second) to the information interaction module 104. After the computer with a touch screen of the information interaction module 104 receives the information, it displays it on the screen in a graphical interface, such as displaying information such as the user name, item name, sterilization start time, and estimated end time on the screen. Among them, the sterilization start time and the estimated end time can also be replaced with the sterilization end countdown.

[0195] If no matching item category is found (including no items in the transfer window, which also belongs to not finding a matching item category), or the item is not allowed to be transferred, or (the value of the allow transfer flag field in the database is "no"), the central control module 105 sends an instruction to open the electromagnetic lock of the transfer window to the electrical control module 102. The instruction format is similar to the operation signal for opening the electromagnetic lock, including two confirmation instructions and one response instruction. The electrical control module 102 receives the instruction and performs the operation of opening the electromagnetic lock.

[0196] At the same time, the central control module 105 records the item matching failure information and sends an alarm signal to the information interaction module 104.

[0197] In one embodiment, the format of the alarm signal is a string, such as "The item information does not match. Please remove the item and operate again". After the computer with a touch screen of the information interaction module 104 receives the alarm signal, an alarm box pops up on the screen to remind the user to remove the corresponding item, and the system returns to step three to wait for new items to be placed and operated.

[0198] Step 5: The electrical control module 102 obtains the sterilization time according to the sterilization status information. When the sterilization time reaches the preset sterilization time, it outputs a sterilization stop signal to stop sterilization.

[0199] In one embodiment, after the electrical control module 102 issues an output sterilization signal to start the sterilization equipment, it starts timing. The timing uses the internal timer of the STM32F103 C8T6 main control chip, and the timer accuracy is 1 millisecond. When the timing reaches the preset sterilization time (such as 60 seconds) in the sterilization instruction, it outputs a sterilization stop signal to stop the operation of the sterilization equipment.

[0200] In one embodiment, the sterilization stop signal format is a fixed frame header (2 bytes), an instruction type (1 byte, with a value of 0x0A indicating the end of sterilization), and a check bit (1 byte, generated using the CRC-8 check algorithm), and this signal is sent to the central control module 105. After receiving the sterilization stop signal, the central control module 105 can choose to store the complete record of this transfer window operation (including user identity information, item classification information, sterilization start time, end time, operation result, etc.) in the information storage module 106 to complete a transfer window operation process.

[0201] Embodiment 3:

[0202] A computer program product includes a computer program, and when the computer program is executed by a processor, it implements the steps of the above method.

[0203] The content not described in detail in this specification belongs to the prior art well-known to those skilled in the art.

Claims

1. An interactive system for a transfer window, characterized in that, Including: The central control module is used to determine whether the identity information of the user matches the pre-stored identity information of the personnel using the transfer window; The electrical control module controls the opening of the electromagnetic lock of the transfer window when the identity information matches. After an item is placed in the transfer window, the electromagnetic lock of the transfer window is closed; The image acquisition module is used to acquire the image data of the item after the electrical control module detects that the electromagnetic lock of the transfer window is closed; The image recognition module classifies the item image data by using an image recognition model obtained by training based on a multi-layer convolutional neural network to obtain item classification information; The central control module is used to determine whether the item classification information matches the pre-stored item type information. When the item information matches, the electrical control module starts sterilization, and the central control module obtains the sterilization status information; The electrical control module is used to obtain the sterilization time according to the sterilization status information. When the sterilization time reaches the preset sterilization time, the sterilization is stopped.

2. The interactive system of the transfer window according to claim 1, characterized in that, In the central control module, when the identity information of the user does not match the pre-stored identity information of the personnel using the transfer window, the central control module records the information indicating the failure of the personnel identity information matching; When the item classification information does not match the pre-stored item type information, the electrical control module opens the electromagnetic lock of the transfer window, and the central control module records the information indicating the failure of the item matching.

3. The interactive system of the transfer window according to claim 1, wherein When the identity information of the user matches the pre-stored identity information of the personnel using the transfer window, the central control module sends an instruction to open the electromagnetic lock of the transfer window to the electrical control module. After receiving the instruction to open the electromagnetic lock of the transfer window, the electrical control module controls the opening of the electromagnetic lock of the transfer window. After an item is placed in the transfer window, the electromagnetic lock of the transfer window is closed; When the electrical control module detects that the electromagnetic lock of the transfer window is closed, the electrical control module sends a feedback instruction for the transfer window to be closed to the central control module. After receiving the feedback instruction for the transfer window to be closed, the central control module sends an image acquisition instruction to the image acquisition module. After receiving the image acquisition instruction, the image acquisition module acquires the item image data. The central control module sends the item image data to the image recognition module, and the image recognition module classifies and recognizes the item image data by using an image recognition model obtained by training based on a multi-layer convolutional neural network to obtain item classification information; The central control module matches the item classification information with the pre-stored item type information to determine whether the item classification information matches the pre-stored item type information: If the item information matches, the central control module sends a sterilization instruction to the electrical control module. After receiving the sterilization instruction, the electrical control module starts the sterilization equipment and sends a sterilization feedback instruction to the central control module. After receiving the sterilization feedback instruction, the central control module obtains the identity information of the user, the item classification information and the sterilization status information. The electrical control module obtains the sterilization time according to the sterilization status information. When the sterilization time reaches the preset sterilization time, a signal to stop sterilization is output to stop the sterilization equipment; the current identity information of the user, the item classification information and the sterilization status information are used for information display; If the item information does not match, the central control module sends an instruction to open the electromagnetic lock of the transfer window to the electrical control module and records the information indicating the failure of the item matching.

4. The interactive system of the transfer window according to claim 3, characterized in that, When the central control module sends an instruction to open the electromagnetic lock of the transfer window, the central control module and the electrical control module interact with each other through a preset double-confirmation protocol, including a confirmation instruction, a response instruction, and an instruction verification mechanism. The instruction interaction method is as follows: The central control module sends the first confirmation instruction to open the electromagnetic lock of the transfer window to the electrical control module. After receiving it, the electrical control module sends a response instruction, and the central control module uses the instruction verification mechanism to parse the response instruction: If the parsing fails or the response instruction is not received, the instruction interaction is aborted; If the parsing is successful, the central control module normally sends the second confirmation instruction to open the electromagnetic lock of the transfer window; After receiving the second confirmation instruction to open the electromagnetic lock of the transfer window, the electrical control module opens the electromagnetic lock of the transfer window.

5. The interactive system of the transfer window according to claim 4, characterized in that, In the electrical control module, the sterilization feedback instruction includes a sterilization start response signal, a sterilization process synchronization signal, and a sterilization process error signal, which are specifically as follows: For the sterilization start response signal sent by the electrical control module, the central control module parses the sterilization start response signal: If the parsing of the sterilization start response signal fails, the sending is terminated; If the parsing of the sterilization start response signal is successful, a process synchronization signal is sent; The electrical control module periodically sends the sterilization process synchronization signal to the central control module through a preset synchronization information sending period. The process synchronization signal uses a timestamp mechanism based on the system time; the system is the operating system or software system of the central control module; When the electrical control module executes the sterilization instruction, it detects whether there is an abnormality in the sterilization process through the transfer window door state feedback or power-off time feedback. If an abnormality is detected, the electrical control module sends a sterilization process error signal to the central control module. After receiving the sterilization process error signal, the central control module obtains the abnormality information and sends it to the information storage module for storage.

6. The interactive system of the transfer window according to claim 5, characterized in that The formats of the sterilization instruction, the first confirmation instruction, the second confirmation instruction, and the transfer window closing feedback instruction are a fixed frame header, an instruction type, and a check bit; the formats of the response instruction and the response signal are a fixed frame header, an instruction type, a response result, and a check bit; The format of the image acquisition instruction is a fixed frame header, an instruction type, acquisition parameters, and a check bit; In the sterilization instruction, the fixed frame header is used to indicate the start of communication, the instruction type is used to indicate the start of sterilization, and the check bit is used to verify the sterilization instruction and ensure the integrity of the communication transmission; In the first confirmation instruction, the fixed frame header is used for communication start, the instruction type is used to indicate a request to open the electromagnetic lock, and the check bit is used to verify the first confirmation instruction and ensure the integrity of the communication transmission; In the response instruction, the fixed frame header is used to indicate the start of communication, the instruction type is used to indicate the response to the request to open the electromagnetic lock, the response result is used to indicate that the request to open the electromagnetic lock has been successfully received, and the check bit is used to verify the response instruction and ensure the integrity of the communication transmission; In the second confirmation instruction, the fixed frame header is used to indicate the start of communication, the instruction type is used to confirm the opening of the electromagnetic lock, and the check bit is used to verify the second confirmation instruction and ensure the integrity of the communication transmission; In the transfer window closing feedback instruction, the fixed frame header is used to indicate the start of communication, the instruction type is used to indicate the window door closing feedback, and the check bit is used to verify the transfer window closing feedback instruction and ensure the integrity of communication transmission; In the image acquisition instruction, the fixed frame header is used to indicate the start of communication, the instruction type is used to indicate the image acquisition request, the acquisition parameters are used to set the resolution and frame rate, and the check bit is used for the image acquisition instruction and ensures the integrity of communication transmission; In the response signal, the fixed frame header is used to indicate the start of communication, the instruction type is the response to the sterilization instruction, the response result is used to indicate the successful reception of the sterilization instruction, and the check bit is used for the response signal and ensures the integrity of communication transmission; The instruction verification mechanism adopts the cyclic redundancy check algorithm.

7. The interactive system of the transfer window according to claim 3, characterized in that In the image recognition module, the method for obtaining the item classification information is as follows: Perform preprocessing on the item image data in the transfer window by first resizing the image and then normalizing it to obtain the preprocessed item image data; The specific method for resizing the item image is: use the bilinear interpolation algorithm to resize all the item image data in the transfer window to the same pixel size; The formula for the normalization operation is: where x normalized represents the normalized x, where x is the image of the item in the transfer window, mean is the brightness mean of the training image dataset of the item image, and std is the brightness standard deviation of the training image dataset of the item image; the training image dataset of the item image is the labeled historical medical supply image dataset; Use the image recognition model trained based on the multi-layer convolutional neural network combined with the pre-trained parameters to process the preprocessed item image data, and obtain the bounding box, item category, and confidence of the preprocessed item image data. The process is as follows: In the convolutional layer of the image recognition model, the convolutional kernel slides and scans the preprocessed item image data to extract local features, obtains the result of the convolutional operation of the local area, takes the result of the convolutional operation of the local area as the output of the neural node of the corresponding area, and obtains the feature map. The expression formula of the convolutional layer is: z(u, v) = Σ i Σ j x i,j × k u-i,v-j + b; where x is the preprocessed object image, and x i,j is the pixel value of the preprocessed object image at the position (i, j). (u, v) represents the position coordinates currently calculated on the output feature map. z(u, v) represents the convolution result obtained at the position (u, v) after applying the convolution kernel to the object image. k is a convolution kernel of size n, which is an n×n matrix, and b is the bias term of the convolutional layer; By calculating the convolutional results at all positions, obtain the feature map z. The feature map z is a matrix of w×w. The calculation formula is: Among them, m is the rank of the x matrix, p is the zero-padding value, and s is the stride of the convolutional kernel; Perform non-linear transformation on the feature map, and then perform downsampling on the non-linearly transformed feature map through the pooling layer. The pooling layer uses average pooling or max pooling for downsampling operations to obtain the pooled feature matrix. The pooled feature matrix undergoes feature integration by the fully connected layer and outputs the feature vector of the bounding box position, item category, and its confidence. The expression of the vector output after the feature integration of the fully connected layer is: y = Wx′ + b1; Among them, x′ is the pooled feature matrix, y is the output vector, that is, the bounding box, item category, and confidence of the item image data, W is the weight matrix, and b1 is the bias term of the fully connected layer; Among them, the bounding box position is represented by pixel coordinates [x1, y1, x2, y2]. x1 and y1 represent the upper left coordinates of the item bounding box, and x2 and y2 represent the lower right coordinates of the item bounding box; the item category corresponds to the category label of the item; the value range of the confidence is 0-1, indicating the credibility of the recognition result; Take the bounding box, item category, and confidence of the item image data as the item classification information.

8. The interactive system of the transfer window according to claim 7, characterized in that, The method for using the image recognition model trained based on the multi-layer convolutional neural network and the pre-trained parameters is: Preprocess the labeled historical medical supply image dataset, divide the preprocessed historical medical supply image dataset into a training set and a validation set, use the training set data to train a multi-layer convolutional neural network model, define a loss function and randomly initialize all weight parameters in the multi-layer convolutional neural network model, and then use the data in the validation set to calculate the gradient of the loss function L(θ) with respect to all weight parameters θ. Calculate the gradient layer by layer through backpropagation. The gradient calculation formula is as follows: where \(L(\theta)\) represents the loss function of the multi-layer convolutional neural network model under a set of \(\theta\). The vector represents the gradient of the loss function \(L(\theta)\) with respect to all weight parameters \(\theta\), and the direction is the direction in which the loss function \(L(\theta)\) grows fastest at \(\theta\). \(\theta\) represents all weight parameters, expressed as \(\theta = (\theta_1,\theta_2,\ldots,\theta\) n ). It is expressed as the partial derivatives of the function \(L(\theta)\) with respect to the parameters \(\theta_1,\theta_2,\ldots,\theta\) n . According to the gradient change and the custom learning rate, use the gradient descent method to update all weight parameters. The parameter update formula is as follows: where η is the learning rate, θ t+1 and θ t represent all the weight parameters of the multi-layer convolutional neural network model at the (t + 1)-th and t-th iterations respectively; L(θ t ) represents the loss function of the multi-layer convolutional neural network model under θ t ; represents the vector of the gradient of the loss function L(θ t ) with respect to all the weight parameters θ; Calculate the loss function value of the validation set after each update. When the change amplitude of the loss function value of the validation set is less than the set threshold or the accuracy of the validation set no longer improves, the training ends. Save the network structure of the multi-layer convolutional neural network and a set of weight parameters with the best performance on the validation set as the image recognition model and pre-training parameters obtained by training based on the multi-layer convolutional neural network; The image recognition model obtained by training based on the multi-layer convolutional neural network includes multiple groups of alternating convolutional layers and pooling layers, and a fully connected layer at the end. The image recognition model obtained by training based on the multi-layer convolutional neural network also includes the size n, stride s, and zero-padding value p of the convolution kernel k in the convolutional layer, the activation function selected by the convolutional layer, the pooling method used by the pooling layer, and the input and output vector dimensions of the convolutional layer and the fully connected layer; The pre-training parameters include the weight matrix W of the fully connected layer, the bias term b1, and the bias terms b of all convolutional layers.

9. A control method for a transfer window, characterized in that, It includes the following steps: Judge whether the identity information of the user is the same as the pre-stored identity information of the personnel using the transfer window; When the identity information is the same, the electrical control module controls the electromagnetic lock of the transfer window to open. When an item is placed in the transfer window, the electromagnetic lock of the transfer window closes; After the electrical control module detects that the electromagnetic lock of the transfer window is closed, the image acquisition module acquires the image data of the item, and the image recognition module uses the image recognition model obtained by training based on the multi-layer convolutional neural network to classify the item image data to obtain the item classification information; Judge whether the item classification information is the same as the pre-stored item type information; When the item information is the same, the electrical control module starts sterilization, and the central control module obtains the sterilization status information; The electrical control module obtains the sterilization time according to the sterilization status information. When the sterilization time reaches the preset sterilization time, the sterilization is stopped.

10. A computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method described in claim 9.

Images