Method and system for detecting cleaning compliance of hotel cleaning personnel based on RFID and deep learning, and electronic equipment

Abstract

The invention discloses a hotel cleaning personnel cleaning compliance detection method and system based on RFID and deep learning, and electronic equipment, an action detection model based on a machine learning model is supplemented and added on the basis of the technology of a detection mechanism based on RFID tag type matching, and the detection result of the model is combined to detect the cleaning compliance of hotel cleaning personnel. And the operation normalization of the cleaning personnel is jointly judged, so that the detection and judgment are more accurate. In addition, a traditional behavior recognition method based on RFID data is complex in model structure, too many in neural network layer number and large in model parameter quantity, the calculation and storage overhead is increased, the model does not have the rapid deployment and response capacity, and in the application scene of the method, efficient operation on equipment with limited calculation resources such as a mobile phone cannot be achieved. The cleaning action recognition module based on the deep learning algorithm recognizes whether cleaning personnel do cleaning actions such as wiping while performing RFID tag type matching detection, so that the detection accuracy and rationality of the system are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of hotel cleaning management, and in particular relates to a cleaning compliance detection method, system and electronic equipment for hotel cleaning staff based on RFID and deep learning. Background Art

[0002] With the continuous development of social economy, travel and hotel accommodation have become more and more popular consumer behaviors.

[0003] As the hygiene requirements of the hotel industry increase, it is particularly important to ensure that the cleaning of each room meets the hygiene standards. Traditional monitoring methods, such as video monitoring, can fully monitor the cleaning process, but this method may infringe on the privacy of cleaning staff and guests, especially in areas with strong privacy such as hotel bathrooms, the application of video monitoring is greatly limited.

[0004] To solve this problem, it is urgent to find a non-visual cleaning detection method. RFID technology is a low-cost and easy-to-use IoT technology that is widely used in asset management, data collection and other scenarios. By deploying RFID tags and using RFID readers to collect RFID tag data, it is possible to effectively detect whether the cleaning staff has cleaned the corresponding area and the use of cleaning tools, thereby avoiding the mixing of rags and other tools in different areas. However, cleaning behavior detection based on pure RFID data has the limitation of low accuracy. Summary of the invention

[0005] To solve the above problems, the purpose of the present invention is to provide a method, system and electronic device for detecting the cleaning compliance of hotel cleaning staff based on RFID and deep learning. The cleaning action recognition module based on the deep learning algorithm can identify whether the cleaning staff is doing cleaning actions such as wiping while performing RFID tag type matching detection, thereby improving the accuracy and rationality of system detection.

[0006] The technical solution provided by the present invention is: a cleaning compliance detection method for hotel cleaning staff based on RFID and deep learning, comprising the following steps:

[0007] Read all static device RFID tags and cleaning tool RFID tags within the target range, where the cleaning tool RFID tags have a unique mapping matching relationship with the static device RFID tags;

[0008] Read the phase data of the dynamic RFID tag configured on the arm of the cleaning staff within the target range, pre-process the phase data using the Gram angular field method to convert it into two-dimensional image data, extract features from the two-dimensional image data through the MobileNetV3 network to generate a feature map, and further extract features from the feature map through the deep separable convolution and attention mechanism to increase the number of channels and expand the perception field of view;

[0009] Through the deep learning-based action detection model, it is determined whether the cleaning staff is performing effective cleaning actions and the detection results are output. According to the label matching mechanism and the cleaning action detection results, a preset warning is made based on preset rules. The detection results include whether the cleaning operations of the cleaning staff comply with the specifications.

[0010] Preferably, reading the phase data of the dynamic RFID tag within the target range and preprocessing the phase data to convert it into two-dimensional image data further comprises:

[0011] The RFID reader is arranged on the cleaning staff's body side, and the RFID reader collects the phase data of the dynamic RFID tag in real time to generate time series data, and the time series data is segmented using a sliding time window and associated with the corresponding dynamic RFID tag;

[0012] The original phase signal data is converted into image data through the Gramian Angular Field method (GAF) as the input data of the deep learning model to reduce the influence of abnormal points caused by interference in the phase signal and obtain richer contextual information in the phase signal, so that the model can better locate and identify the target. At the same time, the image data can make full use of the performance of the deep learning model and improve the recognition accuracy.

[0013] Preferably, performing feature extraction on the two-dimensional image data through a MobileNetV3 network to generate a feature map, and further performing feature extraction on the feature map through a deep separable convolution and attention mechanism further comprises:

[0014] The depthwise separable convolution is configured to decompose the standard convolution into depthwise convolution and point convolution. The depthwise convolution performs convolution processing on the input two-dimensional image data channel by channel, and the point convolution fuses the channel features through a 1×1 convolution kernel, so that the model reduces the number of parameters and the amount of calculation while maintaining a high feature extraction capability. The calculation amount formula is:

[0015] Calculation amount = H × W × (k 2 ×n+n×m), where H and W represent the height and width of the feature map, k×k is the size of the convolution kernel, n is the number of input channels, and m is the number of output channels.

[0016] Preferably, performing feature extraction on the two-dimensional image data through a MobileNetV3 network to generate a feature map, and further extracting features from the feature map through a deep separable convolution and attention mechanism further comprises:

[0017] The MobileNetV3 network also introduces an inverted residual block, which first expands the dimension of the feature map, then processes it through a depth-wise separable convolution, and finally compresses it back to the original dimension.

[0018] Preferably, the method further comprises:

[0019] The generated feature map is pooled vertically and horizontally to further extract global information and enhance the spatial representation ability of the feature. The vertical pooling extracts global information from each column of the feature map. The calculation formula of the vertical pooling is as follows:

[0020]

[0021] Among them, X ij represents the pixel value of the i-th row and j-th column in the feature map, and W is the width of the feature map;

[0022] Horizontal pooling extracts global information from each row of the feature map. The calculation formula for horizontal pooling is:

[0023]

[0024] Among them, H is the height of the feature map.

[0025] Preferably, after vertical and horizontal pooling, the extracted global information is further processed by an independent fully connected layer to generate vertical and horizontal attention weights, thereby enhancing the model's ability to represent features:

[0026] The global information f obtained by vertical pooling y (X) Generate the vertical attention weight through the fully connected layer, the formula is as follows:

[0027] A y (X) = σ(W y ·f y (X)

[0028] Global information f obtained by horizontal pooling x (X) also generates horizontal attention weights through the fully connected layer, the formula is:

[0029] A x (X) = σ(W x ·f x (X)

[0030] Among them, Wy and W x are the weight matrices of the vertical and horizontal fully connected layers respectively, and σ is the activation function;

[0031] The longitudinal attention weight and the lateral attention weight are applied to the original feature map after being processed by independent fully connected layers, thereby weighting the key areas;

[0032] The weighted feature maps are stacked to form a feature pyramid, combining features at different resolutions.

[0033] Preferably, reading all static device RFID tags and cleaning tool RFID tags within the target range and performing one-to-one mapping and matching between the two further includes:

[0034] When multiple cleaning tool RFID tags are read N times consecutively within a preset time, an early warning reminder is generated;

[0035] When only one cleaning tool RFID tag is read N times in a preset time, if the cleaning tool RFID tag matches the type of the static device RFID tag read, the effectiveness of the cleaning action of the cleaning staff is further judged based on the action detection model, otherwise an early warning reminder is generated.

[0036] Based on the same concept, the present invention also provides a hotel cleaning staff cleaning compliance detection system based on RFID and deep learning, including a static device RFID tag, a cleaning tool RFID tag, a dynamic RFID tag, an RFID reader and a preset terminal:

[0037] The dynamic RFID tag is configured on the arm of the cleaning staff, and the cleaning tool RFID tag has a unique mapping correspondence with one of the static device RFID tags;

[0038] The RFID reader is configured on the cleaning staff's body side to read all static device RFID tags and cleaning tool RFID tags within the target range; read the phase data of dynamic RFID tags within the target range, and send the read RFID data to a preset terminal;

[0039] The preset terminal is used to match the type of the read cleaning tool RFID tag with the static device RFID tag, pre-process the phase data of the read dynamic RFID tag and convert it into two-dimensional image data, and then extract features of the two-dimensional image data through the MobileNetV3 network to generate a feature map, and further extract features of the feature map through the enhanced feature extraction network constructed by the fusion of deep separable convolution and attention mechanism to increase the number of channels and expand the perception field of view; then, determine whether the cleaning staff is performing effective cleaning actions through the action detection model based on deep learning; output the detection results and make preset warnings based on preset rules, and the detection results include whether the cleaning operations of the cleaning staff meet the specifications.

[0040] Based on the same concept, the present invention also provides an electronic device, comprising:

[0041] A memory, the memory being used to store a processing program;

[0042] A processor, wherein when executing the processing program, the processor implements any one of the above-mentioned methods for detecting cleaning compliance of hotel cleaning staff based on RFID and deep learning.

[0043] Based on the same concept, the present invention also provides a readable storage medium, on which a processing program is stored. When the processing program is executed by a processor, any one of the above-mentioned hotel cleaning staff cleaning compliance detection methods based on RFID and deep learning is implemented.

[0044] Due to the adoption of the above technical solution, the present invention has the following advantages and positive effects compared with the prior art:

[0045] The technical solution of the present invention adds an action detection model based on a machine learning model to the technology of the detection mechanism based on RFID tag type matching. By combining the detection results of the action detection model, the cleaning standardization of the cleaning staff is jointly judged, and the judgment is more accurate. In addition, traditional behavior recognition methods based on RFID data all use the original phase signal as the model input, which is too sensitive to outliers and abnormal points in the signal. At present, the model structure based on deep learning is relatively complex, the number of neural network layers is too large, and the model parameters are large, which increases the calculation and storage overhead, making the model unable to be quickly deployed and responded. In the application scenario of the present invention, it cannot run efficiently on devices with limited computing resources such as mobile phones. The action detection model of the present invention constructs a lightweight model structure based on MobileNetV3 to achieve an efficient action recognition solution based on RFID phase information. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The specific embodiments of the present invention are further described in detail below with reference to the accompanying drawings, wherein:

[0047] Figure 1 This is a typical hotel bathroom scene diagram;

[0048] Figure 2 It is a schematic diagram of the device composition of the RFID reader / writer of the present invention;

[0049] Figure 3 This is a flow chart of a cleaning compliance detection method for hotel cleaning staff based on RFID and deep learning of the present invention;

[0050] Figure 4 This is a schematic diagram of RFID type pairing detection;

[0051] Figure 5 Design a schematic diagram for the backbone network;

[0052] Figure 6 This is a schematic diagram of the CA network structure;

[0053] Figure 7 It is the CA network flow chart;

[0054] Figure 8 This is a schematic diagram of the MobileNetV3 neural network architecture;

[0055] Fig. 9 A flow chart for designing the action detection model of the present invention.

[0056] Description of reference numerals:

[0057] 1-Static RFID tag (basin); 2-Static RFID tag (toilet); 3-Cleaning tool RFID tag (basin); 4-Cleaning tool RFID tag (toilet); 5-Dynamic RFID tag; 6-basin; 7-toilet; 8-basin cleaning tool (rag); 9-toilet cleaning tool (rag); 10-RFID reader; 11 cleaning staff; 12-cleaning vehicle; 13-sleeves worn by cleaning staff. DETAILED DESCRIPTION

[0058] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The advantages and features of the present invention will become more apparent from the following description and claims. It should be noted that the drawings are all in a very simplified form and use non-precise ratios, which are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention.

[0059] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0060] The application scenario to which the present invention is applicable is to detect the cleaning operations and results of personnel in the main area of ​​the hotel cleaning scene - the bathroom. The schematic diagram of a typical hotel bathroom scene is as follows: Figure 1 As shown. RFID tags with corresponding IDs written in advance are pasted on the basin and toilet; washable flexible RFID tags with corresponding IDs written in advance are sewn on the two rags for wiping the basin and toilet; washable flexible RFID tags with corresponding IDs written in advance are sewn on the sleeves; when cleaning staff perform cleaning operations, they wear sleeves with RFID tags sewn on them and put the RFID reader in their waist bag or pocket. The RFID reader is in working condition during the entire cleaning process.

[0061] The RFID tags used in the system have ID values ​​written in advance, and different types can be distinguished by ID. Each RFID tag ID contains 6 bytes in length, and the encoding structure is as follows:

[0062] Byte sequence number Byte 1 Bytes 2-3 Bytes 4-5 Byte 6 definition Tag Type Hotel No. Room Number Tag serial number Example 34 0005 8721 01

[0063] Among them, byte 1 indicates the tag type, which is defined as follows:

[0064] 12: Toilet cleaning rag

[0065] 23: Basin cleaning rag

[0066] 34: Toilet area

[0067] 45: Basin area

[0068] 56: Sleeve label

[0069] Bytes 2-3: Indicates the unique ID number of the hotel (eg 0005 indicates the hotel ID).

[0070] Bytes 4-5: Room number (e.g. 8721 indicates the room number)

[0071] Byte 6: Tag number

[0072] Therefore, the RFID reader can read the RFID tags in the area and obtain the ID value of the tag, so as to know which area the tag is in, what type of rag it is, or whether it is a tag on a sleeve. Based on this information, detailed cleaning process detection and judgment can be carried out.

[0073] The RFID reader / writer in the present invention adopts an ultra-high frequency UHF RFID module. The MCU controls the RFID module to scan radio frequency signals by sending inventory instructions, and reads the RFID tag data within the effective distance in the environment. At the same time, the RFID reader / writer is connected to the user's mobile phone APP (terminal) through a Bluetooth module. The RFID reader / writer sends valid data to the APP, and the APP can control the start and stop of the equipment and related parameter settings. The mobile phone APP uses a deep learning algorithm to further detect and identify the RFID tag data to determine whether the cleaning process of the cleaning staff is qualified or in violation of regulations. The mobile phone APP regularly uploads the results to the web management platform, and the web management platform provides cleaning result query, statistics, analysis and other functions for regulatory departments and hotel managers. The RFID reader / writer integrates components such as buzzers and indicator lights, and uses buzzers to prompt cleaning staff of the test results with different sounds. The RFID reader / writer is powered by a rechargeable lithium battery. See Figure 2 It is a schematic diagram of the system composition of the RFID reader / writer of the present invention.

[0074] Relevant government departments and hotel managers have several clear requirements for cleaning staff's cleaning operations:

[0075] 1. Clean the guest room bathroom in a timely manner, and clean the key areas of the bathroom, mainly the basin and toilet.

[0076] 2. To wipe the basin and toilet, you must use separate rags and do not mix them.

[0077] Since the RFID tags in the surrounding environment that the RFID reader can scan are directly related to the distance between the reader and the tags, the relative position and direction, etc. After the device is set to a certain RF power, the RFID reader has a reading range and can scan the RFID tags within this range to obtain their IDs.

[0078] During normal operation, if the cleaning staff is ready to start cleaning the basin in the bathroom, they will first take the basin rag from the cart, walk to the basin area, and perform cleaning operations such as wiping. At this time, the RFID reader can continuously read the basin tag and the basin rag tag. Each time the RFID reader scans a data packet, there will be both the basin tag and the basin rag tag. At this time, for the toilet tag, it may be scanned or not, depending on the actual spatial layout of the bathroom. For the toilet rag tag, it is placed on the cart at this time. By requiring the cleaning staff to place the cart outside the bathroom door, it can be ensured that the RFID reader cannot scan the toilet rag tag.

[0079] Similarly, when the cleaning staff finishes cleaning the basin, they will exchange for toilet rags and come to the toilet area to perform cleaning operations such as wiping. At this time, the RFID reader can continuously read the toilet tag and toilet rag tag. For the basin tag, depending on the actual spatial layout of the bathroom, it may be scanned (when the toilet and basin are close to each other), or it may not be scanned. For the basin rag tag, it is placed on the trolley outside the door at this time, and the RFID reader cannot scan it.

[0080] There is no requirement for the order in which cleaning staff clean the basin and toilet.

[0081] Therefore, based on whether the RFID reader can scan the basin / toilet tag, it can be inferred whether the cleaning staff is in the basin / toilet area. Further, based on the type of rag tag scanned by the RFID reader, it can be judged whether the cleaning staff is using a suitable rag for cleaning.

[0082] First embodiment

[0083] The application scenario to which the present invention is applicable is to detect the cleaning operations and results of personnel in the main area of ​​the hotel cleaning scene - the bathroom. The schematic diagram of a typical hotel bathroom scene is as follows: Figure 1 See Figure 3 , which is a flow chart of the cleaning compliance detection method for hotel cleaning staff based on RFID and deep learning in the present invention.

[0084] Before cleaning the guest room bathroom, the cleaning staff turns on the RFID reader device, connects to the reader device through the mobile phone APP, and issues a start-of-work instruction. The RFID reader starts scanning the tag at a fixed cycle (0.1 seconds). The cleaning staff puts the reader device in a waist bag or pocket and keeps the reader with them during the cleaning process.

[0085] The RFID reader periodically scans RFID tags. A certain scanning operation may read one or more types of tags, such as toilet tags, basin tags, tags sewn on sleeves, and two types of rag tags deployed in the bathroom.

[0086] RFID transmits each scan result to the mobile phone APP via Bluetooth. Specifically, for basin tags, toilet tags, basin rag tags, and toilet rag tags, the data sent each time is the tag ID. For the scanned sleeve tag, the tag ID and the phase (angle) data returned by the tag are sent to the mobile phone APP.

[0087] The mobile phone APP performs tag type pairing detection and cleaning action identification based on the received RFID tag data.

[0088] The APP matches the detection results and action recognition results according to the tag type, and determines whether the cleaning staff has completed the cleaning of the basin, whether the toilet, and whether the cleaning process is compliant (that is, whether there is any mixing of rags). The result is sent to the device for prompting, and the result data is reported to the business platform server for storage and statistical analysis.

[0089] The technical solution of this embodiment is: a cleaning compliance detection method for hotel cleaning staff based on RFID and deep learning, comprising the following steps:

[0090] Read all static device RFID tags and cleaning tool RFID tags within the target range, and match the read cleaning tool RFID tags with the static device RFID tags, wherein each cleaning tool RFID tag has a unique mapping correspondence with one of the static device RFID tags;

[0091] The phase data of the dynamic RFID tag within the target range is read, and after the phase data is pre-processed and converted into two-dimensional image data, a motion detection model based on deep learning is used to determine whether the cleaning staff is performing effective cleaning actions, wherein the dynamic RFID tag is configured on the cleaning staff's arm, and the motion detection model based on deep learning:

[0092] The two-dimensional image data is subjected to feature extraction through the MobileNetV3 network to generate a feature map, and the feature map is subjected to further feature extraction through an enhanced feature extraction network constructed by fusing a deep separable convolution and an attention mechanism, so as to increase the number of channels and expand the perception field of view;

[0093] Output the test results and make preset warnings based on preset rules. The test results include whether the cleaning operations of the cleaning staff meet the specifications;

[0094] Among them, the RFID reader / writer for reading the RFID tag is configured on the side of the cleaning staff's body, the RFID reader / writer is connected to a preset terminal signal, and the deep learning-based motion detection model is configured in the terminal.

[0095] The technical solution of this embodiment supplements the technology of the detection mechanism based on RFID tag type matching with an action detection model based on a machine learning model. By combining the detection results of the action detection model, the standardization of the cleaning staff's actions is jointly judged, and the judgment is more accurate. In addition, traditional behavior recognition methods based on RFID data all use the original phase signal as the model input, which is too sensitive to outliers and abnormal points in the signal. At present, the model structure based on deep learning is relatively complex, the number of neural network layers is too large, and the model parameters are large, which increases the calculation and storage overhead, making the model unable to be quickly deployed and responded. In the application scenario of the present invention, it cannot run efficiently on devices with limited computing resources such as mobile phones. The action detection model of the present invention constructs a lightweight model structure based on MobileNetV3 to achieve an efficient action recognition scheme based on RFID phase information. See the figure for the backbone network design.

[0096] Preferably, reading the phase data of the dynamic RFID tag within the target range and preprocessing the phase data to convert it into two-dimensional image data further comprises:

[0097] The RFID reader is arranged on the cleaning staff's body side, and the RFID reader collects the phase data of the dynamic RFID tag in real time to generate time series data, and the time series data is segmented using a sliding time window and associated with the corresponding dynamic RFID tag;

[0098] The original phase signal data is converted into image data through the Gramian Angular Field method (GAF) as the input data of the deep learning model to reduce the influence of abnormal points caused by interference in the phase signal and obtain richer contextual information in the phase signal, so that the model can better locate and identify the target. At the same time, the image data can make full use of the performance of the deep learning model and improve the recognition accuracy.

[0099] The recognition model outputs a detection result and makes a preset warning based on a preset rule, wherein the detection result includes whether it meets the specifications of up and down wiping, wiping in a circle, standing still, and random undefined actions.

[0100] The following is a detailed description of the real-time acquisition of phase data of the cleaning staff's sleeve tag and preprocessing of the phase data, specifically including: using an RFID reader fixed on the cleaning staff's waist to collect phase data of hand movements in real time. The acquisition frequency is every 0.2 seconds, recording the four cleaning actions performed by the cleaning staff in the bathroom:

[0101] Wipe up and down: Move your hand up and down in a straight line.

[0102] Wipe in circles: Move your hands in a circular path.

[0103] Still: The hands remain still and do not move.

[0104] Random undefined actions: Hand movements that do not belong to the first three categories.

[0105] The phase data corresponding to these actions are recorded to form a series of time series data. The time series is segmented using a sliding time window (window size 2 seconds, sliding step size 0.5 seconds) and associated with the corresponding action labels for subsequent processing and model training. In order to make full use of the phase time series data, the system uses GAF technology to convert the phase data from a one-dimensional time series into a two-dimensional image. GAF technology is based on the concept of the Gram matrix in linear algebra and can capture dependencies in time series. The Gram matrix is ​​defined as:

[0106]

[0107] Among them, v1~v n are all k-dimensional vectors, and the inner product of each vector can be expressed as:

[0108]

[0109] The Gram matrix of the phase signal is expressed as:

[0110]

[0111] Where x i is the phase value of the i-th unit time in the signal. Then the phase data is converted from the Cartesian coordinate system to the polar coordinate system, and the expression is:

[0112] θi ＝ arccos(x i ), x∈X

[0113]

[0114] Where R is the constant factor of the polar coordinate system spanning the space, r is the x i The unit distance of the i-th unit time in the signal represents the signal size, θ i is the phase angle of the phase signal at time point i. A Gramian matrix that maintains time dependence can be generated through the GAF conversion algorithm. The Gramian angle sum field based on the cosine function and the Gramian angle difference field based on the sine function are defined as follows:

[0115] GASF=Cos(θ i +θ j ), i, j∈[1,n]

[0116] GADF = cos(θ i -θ j ), i, j∈[1,n]

[0117] where θ i Represents the phase angle value at the i-th time point in the time series, θ j Represents the phase angle value at the jth time point in the time series. i and j are the row and column indices in the matrix, ranging from 1 to n, where n is the length of the phase signal sequence, that is, the total number of time points. The size of the two matrices is n*n.

[0118] Preferably, performing feature extraction on the two-dimensional image data through a MobileNetV3 network to generate a feature map, and further extracting features from the feature map through an enhanced feature extraction network constructed by fusion of deep separable convolution and attention mechanism further comprises:

[0119] The depthwise separable convolution is configured to decompose the standard convolution into depthwise convolution and point convolution. The depthwise convolution performs convolution processing on the input two-dimensional image data channel by channel, which can significantly reduce the amount of calculation. The point convolution fuses the channel features through a 1×1 convolution kernel, so that the model reduces the number of parameters and calculation overhead while maintaining a high feature extraction capability. The calculation amount formula is:

[0120] Calculation amount = H × W × (k 2 ×n+n×m), where H and W represent the height and width of the feature map, k×k is the size of the convolution kernel, n is the number of input channels, and m is the number of output channels.

[0121] Preferably, performing feature extraction on the two-dimensional image data through a MobileNetV3 network to generate a feature map, and further extracting features from the feature map through an enhanced feature extraction network constructed by fusion of deep separable convolution and attention mechanism further includes:

[0122] The MobileNetV3 network also introduces an inverted residual block, which first expands the dimension of the feature map, then processes it through a depth-wise separable convolution, and finally compresses it back to the original dimension.

[0123] See also Figure 8 , by first expanding the dimension of the feature map, then processing it through depthwise separable convolution, and finally compressing it back to the original dimension. This design of the inverted residual block can enhance the network's expressiveness without increasing the computational burden and effectively capture richer image features.

[0124] Furthermore, the method further comprises:

[0125] The generated feature map is pooled vertically and horizontally to further extract global information and enhance the spatial representation ability of the feature. The vertical pooling extracts global information from each column of the feature map. The calculation formula of the vertical pooling is as follows:

[0126]

[0127] Among them, X ij represents the pixel value of the i-th row and j-th column in the feature map, and W is the width of the feature map;

[0128] Horizontal pooling extracts global information from each row of the feature map. The calculation formula for horizontal pooling is:

[0129]

[0130] Among them, H is the height of the feature map.

[0131] The generated preliminary feature map is pooled vertically and horizontally to further extract global information and enhance the spatial representation ability of the feature. This process enables the model to better capture the long-range dependencies and spatial information in the image by performing pooling operations in the vertical and horizontal dimensions respectively. Figure 6 CA network structure diagram, Figure 7 The CA network flow chart is shown in the figure.

[0132] Through vertical pooling, the model is able to integrate important information in each column of features and capture dependencies in the vertical dimension, which enables the model to not only process local features but also understand the vertical features in the image as a whole. The horizontal pooling operation is similar to vertical pooling, but it focuses on extracting information from the horizontal dimension of the image. In this way, the model is able to aggregate features in each row to capture global dependencies in the horizontal dimension. Through the combination of vertical and horizontal pooling, the model is able to not only extract local features in the image, but also effectively obtain global information, making it perform better in complex scenes. This operation helps to capture spatial contextual information in the image, enhance the spatial expression ability of features, and lay the foundation for the subsequent attention mechanism to generate more accurate feature weights.

[0133] Furthermore, after vertical and horizontal pooling, the extracted global information is further processed through independent fully connected layers to generate vertical and horizontal attention weights, thereby enhancing the model's ability to represent features:

[0134] The global information f obtained by vertical pooling y (X) Generate the vertical attention weight through the fully connected layer, the formula is as follows:

[0135] A y (X) = σ(W y ·fy (X)

[0136] Global information f obtained by horizontal pooling x (X) also generates horizontal attention weights through the fully connected layer, the formula is:

[0137] A x (X) = σ(W x ·f x (X)

[0138] Among them, W y and W x are the weight matrices of the vertical and horizontal fully connected layers respectively, and σ is the activation function;

[0139] The longitudinal attention weight and the lateral attention weight are applied to the original feature map after being processed by independent fully connected layers, thereby weighting the key areas;

[0140] The weighted feature maps are stacked to form a feature pyramid, combining features at different resolutions to ensure that the model can maintain high recognition accuracy when processing both large-scale and small-scale objects.

[0141] The feature pyramid is a structure used for multi-scale feature fusion. It can combine features at different resolutions to ensure that the model can maintain high recognition accuracy when processing large-scale and small-scale objects at the same time. Through this feature pyramid structure, the model can effectively deal with objects of different sizes and enhance its robustness in different scenarios. The multi-scale fusion strategy of the feature pyramid can not only capture detailed features, but also help the model understand the complex structure in the image from a global perspective through information interaction at different scales.

[0142] Finally, after the feature pyramid fusion process, the model further improves the ability to capture important spatial location information in the image without significantly increasing the amount of calculation. This design enables the model to perform well in complex scenes while maintaining computational efficiency, and is particularly suitable for use in embedded devices or mobile devices with limited computing resources. The neural network architecture in this design is as follows Figure 8 shown.

[0143] In addition, a type matching detection mechanism is designed in the APP to analyze and judge the RFID tags scanned in real time and output the cleaning results of the cleaning staff.

[0144] The following rule matrix is ​​first defined in the APP:

[0145]

[0146] In the table, a value of "1" indicates that a data packet contains a subtype label, a value of "0" indicates that this type of label is not contained, and a value of "X" indicates any.

[0147] The following is a description of each item in the rule matrix:

[0148] Rule 1: The basin and basin rag are scanned, but the toilet rag is not scanned, which means that the area is in the basin area and is considered normal;

[0149] Rule 2: If the toilet and toilet rag are scanned, but the basin rag is not scanned, it means that the area is in the toilet area and it is judged as normal;

[0150] Rule 3: If a toilet rag and a basin rag are scanned at the same time, and according to the hotel cleaning requirements, two rags cannot be used at the same time, it is considered abnormal;

[0151] Rule 4: The basin and toilet rag are scanned, but the basin rag is not scanned, which means that the user is in the basin area but using the toilet rag, which is considered abnormal.

[0152] Rule 5: If the toilet and basin rag are scanned, but the toilet rag is not scanned, it means that the user is in the toilet area but using the basin rag, which is considered abnormal.

[0153] Rule 6: In other cases, such as only scanning basin / toilet rags, etc., the data will be judged as invalid.

[0154] Preferably, reading all static device RFID tags and cleaning tool RFID tags within the target range, and matching the read cleaning tool RFID tags with the static device RFID tags further comprises:

[0155] When multiple cleaning tool RFID tags are read N times consecutively within a preset time, an early warning reminder is generated;

[0156] When only one cleaning tool RFID tag is read N times in succession within a preset time, if the cleaning tool RFID tag matches the type of at least one static device RFID tag read, the effectiveness of the cleaning action of the cleaning staff is further judged based on the action detection model, otherwise an early warning reminder is generated.

[0157] See also Figure 4 Based on this rule matrix, the APP analyzes the tag data scanned by the reader in real time. When a valid tag is scanned (that is, the basin tag or toilet tag deployed in the system), the room number in the ID is obtained and recorded.

[0158] The data status of each data is marked according to the judgment rules. When any illegal operation is detected, the room cleaning process is marked as illegal and the result is recorded.

[0159] In the absence of any violation, normal data will be detected and counted for the basin and toilet respectively. If the data is normal for 5 consecutive times, the area is considered to be cleaned and recorded in the room cleaning results. In addition to the label matching number rule, the qualified detection of basins and toilets will also be combined with the action recognition output results of the deep learning detection algorithm module. It is necessary to meet the requirement that effective cleaning actions are detected during the label detection process in order to be judged as qualified.

[0160] See also Fig. 9 Before obtaining the cuff phase data of the cleaning personnel in the area to be detected in real time during the cleaning action, the recognition model is also trained. The training method further includes obtaining the cuff phase data of the cleaning personnel to be detected during the cleaning action; dividing the data set into two parts, including a training set and a test set, with a ratio of 7:3, and dividing them into two folders for generating different directories; training the recognition model through the images in the training set, and verifying whether the effect of the training can reach the preset accuracy rate through the images in the test set.

[0161] Second embodiment

[0162] Based on the same concept, the present invention also provides a hotel cleaning staff cleaning compliance detection system based on RFID and deep learning, including a static device RFID tag, a cleaning tool RFID tag, a dynamic RFID tag, an RFID reader, and a preset terminal:

[0163] The RFID reader / writer reads all static device RFID tags and cleaning tool RFID tags within the target range, wherein each cleaning tool RFID tag has a unique mapping correspondence with one of the static device RFID tags; reads the phase data of the dynamic RFID tags within the target range, and sends the read RFID data to a preset terminal;

[0164] The preset terminal matches the type of the cleaning tool RFID tag read with the static device RFID tag, pre-processes the phase data of the dynamic RFID tag read and converts it into two-dimensional image data, then extracts features from the two-dimensional image data through the MobileNetV3 network to generate a feature map, and further extracts features from the feature map through an enhanced feature extraction network constructed by the fusion of deep separable convolution and attention mechanism to increase the number of channels and expand the perception field of view; then, the action detection model based on deep learning is used to determine whether the cleaning staff is performing effective cleaning actions, output the detection results, and make preset warnings based on preset rules, and the detection results include whether the cleaning operation of the cleaning staff meets the specifications;

[0165] Among them, the RFID reader / writer for reading the RFID tag is configured on the side of the cleaning staff's body, the dynamic RFID tag is configured on the cleaning staff's arm, the RFID reader / writer is connected to the preset terminal signal, and the deep learning-based motion detection model is configured in the terminal.

[0166] Based on the same concept, the present invention also provides an electronic device, comprising: a memory, the memory being used to store a processing program; and a processor, the processor implementing any one of the above-mentioned hotel cleaning staff cleaning compliance detection methods based on RFID and deep learning when executing the processing program.

[0167] Based on the same concept, the present invention also provides a readable storage medium, on which a processing program is stored. When the processing program is executed by a processor, any one of the above-mentioned hotel cleaning staff cleaning compliance detection methods based on RFID and deep learning is implemented.

[0168] The cleaning compliance detection method for hotel cleaning staff based on RFID and deep learning can be stored in a computer-readable storage medium if it is implemented in the form of program instructions and sold or used as an independent product. Based on this understanding, the technical solution of this embodiment is essentially or part of the contribution to the prior art or all or part of the technical solution can be embodied in the form of software, and the computer software is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present disclosure. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc. Various media that can store program codes.

[0169] Those skilled in the art can clearly understand that, for the sake of convenience and brevity in description, the identification content specifically executed by the above-described system and device can refer to the corresponding process in the aforementioned method embodiment.

[0170] The embodiments of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments. Even if various changes are made to the present invention, if these changes fall within the scope of the claims of the present invention and their equivalents, they still fall within the protection scope of the present invention.

Claims

1. A cleaning compliance detection method for hotel cleaning staff based on RFID and deep learning, characterized in that: The following steps are involved: Read all static device RFID tags and cleaning tool RFID tags within the target range, where the cleaning tool RFID tags have a unique mapping matching relationship with the static device RFID tags; Read the phase data of the dynamic RFID tag configured on the arm of the cleaning staff within the target range, pre-process the phase data using the Gram angle field to convert it into two-dimensional image data, extract features from the two-dimensional image data through the MobileNetV3 network to generate a feature map, and further extract features from the feature map through the deep separable convolution and attention mechanism to increase the number of channels and expand the perception field of view; Use the deep learning-based action detection model to determine whether the cleaning staff is performing effective cleaning actions and output the detection results; According to the tag matching mechanism and the cleaning action detection results, a preset warning is made based on the preset rules. The detection results include whether the cleaning operation of the cleaning staff complies with the specifications.

2. The cleaning compliance detection method for hotel cleaning staff based on RFID and deep learning according to claim 1 is characterized in that: Reading phase data of a dynamic RFID tag within a target range and preprocessing the phase data to convert it into two-dimensional image data further includes: The RFID reader is arranged on the cleaning staff's body side, and the RFID reader collects the phase data of the dynamic RFID tag in real time to generate time series data, and the time series data is segmented using a sliding time window and associated with the corresponding dynamic RFID tag; The original phase signal data is converted into image data through the Gramian Angular Field method (GAF) as the input data of the deep learning model to reduce the influence of abnormal points caused by interference in the phase signal and obtain richer contextual information in the phase signal, so that the model can better locate and identify the target. At the same time, the image data can make full use of the performance of the deep learning model and improve the recognition accuracy.

3. The cleaning compliance detection method for hotel cleaning staff based on RFID and deep learning according to claim 1 is characterized in that: Performing feature extraction on the two-dimensional image data through the MobileNetV3 network to generate a feature map, and further performing feature extraction on the feature map through a deep separable convolution and attention mechanism further includes: The depthwise separable convolution is configured to decompose the standard convolution into depthwise convolution and point convolution. The depthwise convolution performs convolution processing on the input two-dimensional image data channel by channel, and the point convolution fuses the channel features through a 1×1 convolution kernel, so that the model reduces the number of parameters and the amount of calculation while maintaining a high feature extraction capability. The calculation amount formula is: Calculation amount = H × W × (k 2 ×n+n×m), where H and W represent the height and width of the feature map, k×k is the size of the convolution kernel, n is the number of input channels, and m is the number of output channels.

4. The cleaning compliance detection method for hotel cleaning staff based on RFID and deep learning according to claim 3 is characterized in that: Performing feature extraction on the two-dimensional image data through the MobileNetV3 network to generate a feature map, and further extracting features from the feature map through a deep separable convolution and attention mechanism further includes: The MobileNetV3 network also introduces an inverted residual block, which first expands the dimension of the feature map, then processes it through a depth-wise separable convolution, and finally compresses it back to the original dimension.

5. The cleaning compliance detection method for hotel cleaning staff based on RFID and deep learning according to claim 3 is characterized in that: The method further comprises: The generated feature map is pooled vertically and horizontally to further extract global information and enhance the spatial representation ability of the feature. The vertical pooling extracts global information from each column of the feature map. The calculation formula of the vertical pooling is as follows: Among them, X ij represents the pixel value of the i-th row and j-th column in the feature map, and W is the width of the feature map; Horizontal pooling extracts global information from each row of the feature map. The calculation formula for horizontal pooling is: Among them, H is the height of the feature map.

6. The cleaning compliance detection method for hotel cleaning staff based on RFID and deep learning according to claim 5 is characterized in that: After vertical and horizontal pooling, the extracted global information is further processed through independent fully connected layers to generate vertical and horizontal attention weights, thereby enhancing the model's ability to represent features: The global information f obtained by vertical pooling y (X) Generate the vertical attention weight through the fully connected layer, the formula is as follows: A y (X)=σ(W y ·f y (X)) Global information f obtained by horizontal pooling x (X) also generates horizontal attention weights through the fully connected layer, the formula is: A x (X)=σ(W x ·f x (X)) Among them, W y and W x are the weight matrices of the vertical and horizontal fully connected layers respectively, and σ is the activation function; The longitudinal attention weight and the lateral attention weight are applied to the original feature map after being processed by independent fully connected layers, thereby weighting the key areas; The weighted feature maps are stacked to form a feature pyramid, combining features at different resolutions.

7. The cleaning compliance detection method for hotel cleaning staff based on RFID and deep learning according to claim 1 is characterized in that: Reading all static device RFID tags and cleaning tool RFID tags within the target range and mapping the two one by one further includes: When multiple cleaning tool RFID tags are read N times in succession within a preset time, an early warning reminder is generated; When only one cleaning tool RFID tag is read N times in a preset time, if the cleaning tool RFID tag matches the type of the static device RFID tag read, the effectiveness of the cleaning action of the cleaning staff is further judged based on the action detection model, otherwise an early warning reminder is generated.

8. A hotel cleaning staff cleaning compliance detection system based on RFID and deep learning, characterized in that: Including static device RFID tags, cleaning tool RFID tags, dynamic RFID tags, RFID readers and writers, and preset terminals: The dynamic RFID tag is configured on the arm of the cleaning staff, and the cleaning tool RFID tag has a unique mapping correspondence with one of the static device RFID tags; The RFID reader is configured on the cleaning staff's body side to read all static device RFID tags and cleaning tool RFID tags within the target range; read the phase data of dynamic RFID tags within the target range, and send the read RFID data to a preset terminal; The preset terminal is used to match the type of the read cleaning tool RFID tag with the static device RFID tag, pre-process the phase data of the read dynamic RFID tag and convert it into two-dimensional image data, and then extract features of the two-dimensional image data through the MobileNetV3 network to generate a feature map, and further extract features of the feature map through the enhanced feature extraction network constructed by the fusion of deep separable convolution and attention mechanism to increase the number of channels and expand the perception field of view; then, determine whether the cleaning staff is performing effective cleaning actions through the action detection model based on deep learning; output the detection results and make preset warnings based on preset rules, and the detection results include whether the cleaning operations of the cleaning staff meet the specifications.

9. An electronic device, characterized in that: include: A memory, the memory being used to store a processing program; A processor, wherein when executing the processing program, the processor implements the cleaning compliance detection method for hotel cleaning staff based on RFID and deep learning as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium stores a processing program, and when the processing program is executed by the processor, the cleaning compliance detection method for hotel cleaning staff based on RFID and deep learning as described in any one of claims 1 to 7 is implemented.

Images