IPTV (Internet Protocol Television) set top box quality detection method, device, equipment and medium

Abstract

The invention provides an IPTV set top box quality detection method and device, equipment and a medium, and relates to the technical field of set top box quality detection. The method comprises the following steps: firstly, acquiring set top box image data obtained by performing image acquisition on a target IPTV set top box; secondly, performing feature extraction processing on the set top box image data through a feature extraction model in an information identification network formed by pre-training, and outputting a set top box image semantic vector corresponding to the set top box image data; and then, through a feature recognition model in an information recognition network, in combination with reference set top box image data, performing feature recognition processing on the set top box image semantic vector, and outputting target set top box quality data corresponding to the target IPTV set top box. Based on the above content, the problem that the reliability of the quality detection of the set top box is relatively low in the prior art can be improved.

Description

Technical Field

[0001] This application relates to the technical field of set-top box quality detection. Specifically, it relates to a method, device, equipment and medium for detecting the quality of IPTV set-top boxes. Background Art

[0002] The IPTV (Internet Protocol Television, which uses Internet protocols to transmit TV programs and video content. Different from traditional cable or satellite TV, IPTV uses the Internet as the transmission medium, allowing users to watch TV content through a network connection) set-top boxes formed by injection molding are neatly arranged in the printing jig according to the direction and stored at the position of the automatic loading machine. After starting, the device self-checks whether the direction is incorrect. After passing, the automatic feeder sends the product to the next process. Save the content to be printed in the three-dimensional printing device. This product can be printed in three directions, and this device can perform multi-sided printing at the same time. After printing, it enters the drying area on the production line. In this area, the device sets the temperature and adjusts the transportation speed on the line to determine the baking time. Ensure that the product has sufficient drying time without printing or wet phenomena. After drying, it enters the inspection and supervision area. First, set product information, specifications, printing color, font size, size requirements, position requirements, etc., and save the relevant information according to the product type for convenient retrieval of parameters later. Then, check according to the entered information, automatically import the inspection information, and transfer the qualified products to the next process. The unqualified products are automatically transferred to the automatic erasing and repairing equipment. According to the received unqualified product information, erase and clean the corresponding surface or points to be scrubbed, and complete the self-inspection of the erased part. Re-print the products to be reworked. Therefore, the reliability of the printed information on the products can be guaranteed. In the prior art, generally, relevant information is extracted based on traditional character recognition technologies (such as OCR (Optical Character Recognition)). Then, the extracted relevant information can be compared with the reference information. However, the inventor has found through research that traditional OCR has poor fault tolerance for image noise, low resolution, blur, rotation or tilt and other deformations. When the image quality is poor, traditional OCR may not be able to correctly recognize characters. Therefore, in the prior art, there is an easy problem that the reliability of the quality detection of set-top boxes is relatively low. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a method, device, equipment and medium for detecting the quality of IPTV set-top boxes to improve the problem of relatively low reliability of set-top box quality detection existing in the prior art.

[0004] To achieve the above purpose, this application adopts the following technical solutions:

[0005] An IPTV set-top box quality detection method, comprising:

[0006] Obtaining set-top box image data obtained by performing image acquisition on a target IPTV set-top box, wherein the box body of the target IPTV set-top box has set-top box related information formed by a printing process;

[0007] Performing feature extraction processing on the set-top box image data through a feature extraction model in an information recognition network formed by pre-training, and outputting a set-top box image semantic vector corresponding to the set-top box image data, wherein the information recognition network belongs to a neural network;

[0008] Performing feature recognition processing on the set-top box image semantic vector through a feature recognition model in the information recognition network in combination with reference set-top box image data, and outputting target set-top box quality data corresponding to the target IPTV set-top box, wherein the reference set-top box image data has qualified set-top box related information, and the target set-top box quality data is used to reflect whether the set-top box related information on the box body of the target IPTV set-top box is qualified.

[0009] In a preferred selection of the present application, in the above IPTV set-top box quality detection method, the step of performing feature extraction processing on the set-top box image data through a feature extraction model in an information recognition network formed by pre-training and outputting a corresponding set-top box image semantic vector includes:

[0010] Performing convolution processing on multiple channel data of the set-top box image data respectively through a convolution unit in a feature extraction model in an information recognition network formed by pre-training, and forming multiple channel data convolution vectors corresponding to the set-top box image data, wherein the multiple channel data at least includes red channel data, green channel data, blue channel data and grayscale image channel data;

[0011] Multiplying the channel data convolution vector corresponding to the grayscale image channel data by the channel data convolution vector corresponding to the red channel data to obtain a corresponding correlation parameter distribution, and performing weighted calculation on the channel data convolution vector corresponding to the grayscale image channel data based on the correlation parameter distribution to form a corresponding first correlation mining vector;

[0012] Multiplying the channel data convolution vector corresponding to the grayscale image channel data by the channel data convolution vector corresponding to the green channel data to obtain a corresponding correlation parameter distribution, and performing weighted calculation on the channel data convolution vector corresponding to the grayscale image channel data based on the correlation parameter distribution to form a corresponding second correlation mining vector;

[0013] Multiply the channel data convolution vector corresponding to the grayscale image channel data and the channel data convolution vector corresponding to the blue channel data to obtain the corresponding correlation parameter distribution, and based on this correlation parameter distribution, perform weighted calculation on the channel data convolution vector corresponding to the grayscale image channel data to form the corresponding third correlation mining vector;

[0014] Perform mean calculation or splicing processing on the channel data convolution vector corresponding to the grayscale image channel data, the first correlation mining vector, the second correlation mining vector, and the third correlation mining vector to form the corresponding set-top box image semantic vector.

[0015] In a preferred selection of this application, in the above IPTV set-top box quality detection method, the step of respectively performing convolution processing on multiple channel data of the set-top box image data through the convolution unit in the feature extraction model in the information recognition network formed by pre-training to form multiple channel data convolution vectors corresponding to the set-top box image data includes:

[0016] Through the convolution unit in the feature extraction model in the information recognition network formed by pre-training, respectively perform convolution processing on the red channel data, green channel data, and blue channel data of the set-top box image data to form a channel data convolution vector corresponding to the red channel data, a channel data convolution vector corresponding to the green channel data, and a channel data convolution vector corresponding to the blue channel data;

[0017] Perform binarization processing on the grayscale image channel data of the set-top box image data to obtain the binarized image data corresponding to the set-top box image data;

[0018] Perform at least one morphological combination operation on the binarized image data to obtain at least one morphological image data corresponding to the binarized image data, where each morphological combination operation includes one dilation operation and one erosion operation;

[0019] Perform convolution processing on the grayscale image channel data through the convolution unit to form the grayscale image convolution vector corresponding to the grayscale image channel data, and perform convolution processing on each morphological image data through the convolution unit to form the morphological image convolution vector corresponding to each morphological image data;

[0020] Perform correlation mining processing on the grayscale image convolution vector corresponding to the grayscale image channel data and the morphological image convolution vector corresponding to each morphological image data, and output the channel data convolution vector corresponding to the grayscale image channel data.

[0021] In a preferred option of the present application, in the above IPTV set-top box quality detection method, the step of performing correlation mining processing on the grayscale image convolution vector corresponding to the grayscale image channel data and each morphological image convolution vector corresponding to the morphological image data, and outputting the channel data convolution vector corresponding to the grayscale image channel data includes:

[0022] For each morphological image convolution vector corresponding to the morphological image data, multiply the morphological image convolution vector by the restored image convolution vector corresponding to the grayscale image channel data to obtain a corresponding correlation parameter distribution, and based on the correlation parameter distribution, perform weighted calculation on the restored image convolution vector corresponding to the grayscale image channel data, and output a morphological image correlation mining vector corresponding to the grayscale image channel data;

[0023] Perform mean calculation or splicing processing on each morphological image correlation mining vector corresponding to the grayscale image channel data and the grayscale image convolution vector corresponding to the grayscale image channel data to form the channel data convolution vector corresponding to the grayscale image channel data.

[0024] In a preferred option of the present application, in the above IPTV set-top box quality detection method, the step of performing feature recognition processing on the set-top box image semantic vector through the feature recognition model in the information recognition network and combining the reference set-top box image data, and outputting the target set-top box quality data corresponding to the target IPTV set-top box includes:

[0025] Perform feature extraction processing on the reference set-top box image data through the feature extraction model, and output the reference image semantic vector corresponding to the reference set-top box image data;

[0026] Through the embedding unit included in the feature recognition model in the information recognition network, perform feature embedding processing on the set-top box description text corresponding to the qualified set-top box related information in the reference set-top box image data, and output the description text embedding vector corresponding to the set-top box description text;

[0027] Through the correlation mining unit included in the feature recognition model, perform correlation mining on the reference image semantic vector based on the description text embedding vector, and output the reference image mining vector corresponding to the reference image semantic vector;

[0028] Based on the vector distance between the reference image mining vector and the set-top box image semantic vector, determine the target set-top box quality data corresponding to the target IPTV set-top box, so as to complete the feature recognition processing on the set-top box image semantic vector.

[0029] In a preferred option of the present application, in the above IPTV set-top box quality detection method, the step of performing correlation mining on the reference image semantic vector based on the description text embedding vector by the correlation mining unit included in the feature recognition model, and outputting a reference image mining vector corresponding to the reference image semantic vector includes:

[0030] Performing correlation mining with multiple depths on the reference image semantic vector based on the description text embedding vector by the correlation mining unit included in the feature recognition model, and outputting multiple depth correlation mining vectors corresponding to the reference image semantic vector;

[0031] Performing mean calculation or splicing processing on the multiple depth correlation mining vectors corresponding to the reference image semantic vector, and outputting a reference image mining vector corresponding to the reference image semantic vector.

[0032] In a preferred option of the present application, in the above IPTV set-top box quality detection method, the step of performing correlation mining with multiple depths on the reference image semantic vector based on the description text embedding vector by the correlation mining unit included in the feature recognition model, and outputting multiple depth correlation mining vectors corresponding to the reference image semantic vector includes:

[0033] Loading the description text embedding vector and the reference image semantic vector into the correlation mining unit included in the feature recognition model;

[0034] In the correlation mining of the first depth, multiplying the description text embedding vector and the reference image semantic vector to obtain a corresponding correlation parameter distribution, and performing weighted calculation on the reference image semantic vector based on this correlation parameter distribution, and outputting a first depth mining vector;

[0035] In the correlation mining of the second depth, respectively performing convolution processing on the description text embedding vector and the first depth mining vector to form a second depth convolutional description text embedding vector and a convolutional reference image semantic vector, and multiplying the convolutional description text embedding vector and the convolutional reference image semantic vector to obtain a corresponding correlation parameter distribution, and performing weighted calculation on the convolutional reference image semantic vector based on this correlation parameter distribution, and outputting a second depth mining vector;

[0036] In the correlation mining at each depth after the second depth, the convolutional description text embedding vector and the depth mining vector at the previous depth are respectively subjected to convolutional processing to form the convolutional description text embedding vector and the convolutional reference image semantic vector at the current depth, and, the convolutional description text embedding vector and the convolutional reference image semantic vector are multiplied to obtain the corresponding correlation parameter distribution, and, based on the correlation parameter distribution, weighted calculation is performed on the convolutional reference image semantic vector to output the current depth mining vector;

[0037] Deconvolution processing is respectively performed on the depth mining vectors at each depth after the second depth to obtain a plurality of vectors having the same size as the first depth mining vector, and, the plurality of vectors and the first depth mining vector are used as a plurality of depth correlation mining vectors corresponding to the reference image semantic vector.

[0038] The present application also provides an IPTV set-top box quality detection device, including:

[0039] A set-top box image acquisition module, configured to acquire set-top box image data obtained by performing image acquisition on a target IPTV set-top box, wherein, on the box body of the target IPTV set-top box, there is set-top box related information formed by a printing process;

[0040] A feature extraction module, configured to perform feature extraction processing on the set-top box image data through a feature extraction model in a pre-trained information recognition network, and output a set-top box image semantic vector corresponding to the set-top box image data, wherein, the information recognition network belongs to a neural network;

[0041] A feature recognition module, configured to perform feature recognition processing on the set-top box image semantic vector through a feature recognition model in the information recognition network in combination with reference set-top box image data, and output target set-top box quality data corresponding to the target IPTV set-top box, wherein, in the reference set-top box image data, there is qualified set-top box related information, and the target set-top box quality data is used to reflect whether the set-top box related information on the box body of the target IPTV set-top box is qualified.

[0042] On the above basis, the present application also provides an electronic device, including:

[0043] A memory, configured to store a computer program;

[0044] A processor connected to the memory, configured to execute the computer program stored in the memory to implement the above IPTV set-top box quality detection method.

[0045] On the basis described above, the present application further provides a computer-readable storage medium, in which a computer program is stored, and when the computer program runs, it executes each step of the above IPTV set-top box quality detection method.

[0046] For the IPTV set-top box quality detection method, device, equipment and medium provided by the present application, first, obtain the set-top box image data obtained by image acquisition of the target IPTV set-top box; secondly, perform feature extraction processing on the set-top box image data through the feature extraction model in the information recognition network formed by pre-training, and output the set-top box image semantic vector corresponding to the set-top box image data; then, through the feature recognition model in the information recognition network, in combination with the reference set-top box image data, perform feature recognition processing on the set-top box image semantic vector, and output the target set-top box quality data corresponding to the target IPTV set-top box. Based on the above, on the one hand, the powerful learning ability of the neural network can be utilized to enable reliable feature recognition processing, so as to obtain reliable target set-top box quality data; on the other hand, since the reference set-top box image data is also used as an auxiliary during the feature recognition processing, the reliability of the feature recognition processing is improved. Therefore, it can effectively ensure that the obtained target set-top box quality data can truly and reliably reflect the actual printing situation of the set-top box, thereby improving the relatively low reliability of set-top box quality detection existing in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] To make the above objects, features and advantages of the present application more obvious and understandable, the following specifically enumerates preferred embodiments and, in conjunction with the accompanying drawings, makes a detailed description as follows.

[0048] Figure 1 It is a block diagram of the structure of the electronic device provided by the embodiment of the present application.

[0049] Figure 2 It is a schematic flowchart of the IPTV set-top box quality detection method provided by the embodiment of the present application.

[0050] Figure 3 It is a schematic diagram of the correlation mining of multiple depths provided by the embodiment of the present application.

[0051] Figure 4 It is a schematic block diagram of the IPTV set-top box quality detection device provided by the embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the following will clearly and completely describe the technical solutions in the embodiments of this application with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all of the embodiments. Usually, the components of the embodiments of this application described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations.

[0053] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of this application claimed, but merely represents selected embodiments of this application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in this application without creative efforts fall within the scope of protection of this application.

[0054] As Figure 1 shown, the embodiments of this application provide an electronic device. Among them, the electronic device may include a memory, a processor, and an IPTV set-top box quality detection device.

[0055] Specifically, the memory and the processor are directly or indirectly electrically connected to achieve data transmission or interaction. For example, the memory and the processor can be electrically connected through one or more communication buses or signal lines. The IPTV set-top box quality detection device includes at least one software function module stored in the memory in the form of software or firmware. The processor is used to execute the executable computer programs stored in the memory, such as the software function modules and computer programs included in the IPTV set-top box quality detection device, etc., to implement the IPTV set-top box quality detection method provided by the embodiments of this application.

[0056] Optionally, the memory may be, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc.

[0057] Moreover, the processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), a System on Chip (SoC), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0058] It can be understood that Figure 1 The structure shown is only schematic, and the electronic device may further include more or fewer components than those shown Figure 1 herein, or have a different configuration from that shown Figure 1 For example, it may further include a communication unit for information interaction with other devices (such as an image acquisition device, etc.).

[0059] Combined with Figure 2 this, an IPTV set-top box quality detection method applicable to the above electronic device is further provided in an embodiment of the present application. Among them, the method steps defined by the processes related to the IPTV set-top box quality detection method can be implemented by the electronic device.

[0060] Next, Figure 2 the specific process shown will be elaborated in detail.

[0061] Step S110, obtain the set-top box image data obtained by image acquisition of the target IPTV set-top box.

[0062] In an embodiment of the present application, the electronic device may obtain the set-top box image data obtained by image acquisition of the target IPTV set-top box. For example, it may perform an image acquisition operation on the box body of the target IPTV set-top box through a corresponding image acquisition device. In addition, the set-top box image data may include one frame of image or multiple frames of images, and the multiple frames of images may be processed separately. For example, one frame of image corresponds to one surface of the target IPTV set-top box. Among them, the box body of the target IPTV set-top box has set-top box related information formed by a printing process, and the specific content of the set-top box related information is not limited. For example, it may include product information, specification models, etc.

[0063] Step S120, perform feature extraction processing on the set-top box image data through a feature extraction model in an information recognition network formed by pre-training, and output a set-top box image semantic vector corresponding to the set-top box image data.

[0064] In an embodiment of the present application, after obtaining the set-top box image data, the electronic device may perform feature extraction processing on the set-top box image data through a feature extraction model in an information recognition network formed by pre-training, and output a set-top box image semantic vector corresponding to the set-top box image data. That is to say, the semantic information in the set-top box image data can be mined and represented in the form of a vector. Among them, the information recognition network belongs to a neural network.

[0065] Step S130, through the feature recognition model in the information recognition network, in combination with the reference set-top box image data, perform feature recognition processing on the set-top box image semantic vector, and output target set-top box quality data corresponding to the target IPTV set-top box.

[0066] In an embodiment of the present application, after obtaining the set-top box image semantic vector, the electronic device may perform feature recognition processing on the set-top box image semantic vector through the feature recognition model in the information recognition network, in combination with the reference set-top box image data, and output target set-top box quality data corresponding to the target IPTV set-top box. Among them, the reference set-top box image data has qualified set-top box-related information (for example, an image acquisition operation can be performed on a qualified IPTV set-top box to obtain the corresponding reference set-top box image data), and the target set-top box quality data is used to reflect whether the set-top box-related information on the box body of the target IPTV set-top box is qualified. For example, when the content, color, font, size, position, etc. of the set-top box-related information in the set-top box image data are all consistent with the content, color, font, size, position, etc. of the set-top box-related information in the reference set-top box image data, it is determined to be qualified, otherwise it is determined to be unqualified. That is to say, during the training process of the information recognition network, only when the content, color, font, size, position, etc. of the set-top box-related information in the sample set-top box image data are all consistent with the content, color, font, size, position, etc. of the set-top box-related information in the reference set-top box image data, the corresponding label is qualified, otherwise, the corresponding label is unqualified.

[0067] Based on the above content, on the one hand, the powerful learning ability of the neural network can be utilized to enable reliable feature recognition processing, so as to obtain reliable target set-top box quality data; on the other hand, since the reference set-top box image data is also used as an auxiliary during the feature recognition processing, the reliability of the feature recognition processing is improved. Therefore, it can effectively ensure that the obtained target set-top box quality data can truly and reliably reflect the actual printing situation of the set-top box, thereby improving the problem of relatively low reliability of set-top box quality detection existing in the prior art.

[0068] Regarding step S120, it should be noted that the specific method for extracting features from the set-top box image data is not limited and can be selected according to actual needs.

[0069] For example, in an alternative embodiment, the set-top box image data can be directly subjected to convolution processing, and then the semantic vector obtained by the convolution processing is used as the set-top box image semantic vector corresponding to the set-top box image data. In this way, the efficiency of feature extraction processing can be improved to a certain extent, and the corresponding computational complexity can be reduced.

[0070] Again, for example, in another alternative embodiment, in order to improve the reliability of feature extraction processing and enable the obtained set-top box image semantic vector to fully and reliably represent the semantic information in the set-top box image data, the above step S120 may further include step S121, step S122, step S123, step S124, and step S125, and the specific content is as follows.

[0071] Step S121: Respectively perform convolution processing on the multi-channel data of the set-top box image data through the convolution unit in the feature extraction model in the information recognition network formed by pre-training, to form a plurality of channel data convolution vectors corresponding to the set-top box image data.

[0072] In the embodiment of the present application, the multi-channel data of the set-top box image data can be respectively subjected to convolution processing through the convolution unit in the feature extraction model in the information recognition network formed by pre-training, to form a plurality of channel data convolution vectors corresponding to the set-top box image data. Among them, the multi-channel data at least includes red channel data, green channel data, blue channel data, and grayscale image channel data. In this way, 4 channel data convolution vectors can be obtained. Since there are requirements for the color of the printed characters based on some needs, by separately mining the data of each color channel, more detailed information can be noticed, thereby improving the reliability of subsequent recognition.

[0073] Step S122: Multiply the channel data convolution vector corresponding to the grayscale image channel data by the channel data convolution vector corresponding to the red channel data to obtain a corresponding correlation parameter distribution, and based on this correlation parameter distribution, perform weighted calculation on the channel data convolution vector corresponding to the grayscale image channel data to form a corresponding first correlation mining vector.

[0074] In the embodiments of the present application, after obtaining the corresponding channel data convolution vectors, the channel data convolution vector corresponding to the grayscale image channel data and the channel data convolution vector corresponding to the red channel data can be multiplied to obtain the corresponding correlation parameter distribution, and based on this correlation parameter distribution, the channel data convolution vector corresponding to the grayscale image channel data is weighted and calculated to form the corresponding first correlation mining vector. It should be noted that the channel data convolution vector corresponding to the grayscale image channel data actually belongs to the representation of the semantic information of the entire printed text. In this way, the semantic information corresponding to the red channel data can be fused into the semantic information of the entire printed text, so that when representing the overall semantic information, the semantic information corresponding to the red channel data can also be taken into account, thereby improving the ability of semantic representation. In addition, the correlation parameter distribution can refer to the dot product between the transposed vector of the channel data convolution vector corresponding to the grayscale image channel data and the channel data convolution vector corresponding to the red channel data. The first correlation mining vector can refer to the result of weighted summation calculation of the channel data convolution vector corresponding to the grayscale image channel data based on this dot product. In this way, the association and fusion of different semantic information can be realized.

[0075] Step S123: Multiply the channel data convolution vector corresponding to the grayscale image channel data and the channel data convolution vector corresponding to the green channel data to obtain the corresponding correlation parameter distribution, and based on this correlation parameter distribution, the channel data convolution vector corresponding to the grayscale image channel data is weighted and calculated to form the corresponding second correlation mining vector.

[0076] In the embodiments of the present application, after obtaining the corresponding channel data convolution vectors, the channel data convolution vector corresponding to the grayscale image channel data and the channel data convolution vector corresponding to the green channel data can also be multiplied to obtain the corresponding correlation parameter distribution, and based on this correlation parameter distribution, the channel data convolution vector corresponding to the grayscale image channel data is weighted and calculated to form the corresponding second correlation mining vector. The specific processing process is as described above.

[0077] Step S124: Multiply the channel data convolution vector corresponding to the grayscale image channel data and the channel data convolution vector corresponding to the blue channel data to obtain the corresponding correlation parameter distribution, and based on this correlation parameter distribution, the channel data convolution vector corresponding to the grayscale image channel data is weighted and calculated to form the corresponding third correlation mining vector.

[0078] In an embodiment of the present application, after obtaining the corresponding channel data convolution vectors, the channel data convolution vector corresponding to the grayscale image channel data and the channel data convolution vector corresponding to the blue channel data may also be multiplied to obtain the corresponding correlation parameter distribution, and based on this correlation parameter distribution, a weighted calculation is performed on the channel data convolution vector corresponding to the grayscale image channel data to form the corresponding third correlation mining vector. The specific processing process is as described above.

[0079] Step S125: Perform a mean calculation or splicing process on the channel data convolution vector corresponding to the grayscale image channel data, the first correlation mining vector, the second correlation mining vector, and the third correlation mining vector to form the corresponding set-top box image semantic vector.

[0080] In an embodiment of the present application, after separately performing fusion to obtain the first correlation mining vector, the second correlation mining vector, and the third correlation mining vector, a mean calculation or splicing process may be performed on the channel data convolution vector corresponding to the grayscale image channel data, the first correlation mining vector, the second correlation mining vector, and the third correlation mining vector to form the corresponding set-top box image semantic vector. In this way, the set-top box image semantic vector can pay attention to more important information, thereby improving the representation accuracy of semantic information.

[0081] It can be understood that in the above step S121, the specific manner of separately performing convolution processing on the multiple channel data of the set-top box image data is not limited either, and corresponding selection can be made according to actual needs. For example, in an alternative embodiment, in order to enable the obtained channel data convolution vector to pay more attention to and represent the semantic information of characters, thereby improving the reliability of subsequent processing, the above step S121 may further include step S121a, step S121b, step S121c, step S121d, and step S121e. The specific content is as follows.

[0082] Step S121a: Respectively perform convolution processing on the red channel data, green channel data, and blue channel data of the set-top box image data through the convolution unit in the feature extraction model in the information recognition network formed by pre-training to form the channel data convolution vector corresponding to the red channel data, the channel data convolution vector corresponding to the green channel data, and the channel data convolution vector corresponding to the blue channel data.

[0083] In an embodiment of the present application, the convolution units in the feature extraction model of the information recognition network formed by pre-training can be used to perform convolution processing on the red channel data, green channel data, and blue channel data of the set-top box image data respectively, so as to form a channel data convolution vector corresponding to the red channel data, a channel data convolution vector corresponding to the green channel data, and a channel data convolution vector corresponding to the blue channel data. Exemplarily, considering that the important information represented by different color channel data is different, the convolution unit may include a plurality of convolution sub-units, so that the red channel data, green channel data, and blue channel data can be respectively subjected to convolution processing through the plurality of convolution sub-units. In this way, the accuracy of the convolution processing can be improved.

[0084] Step S121b: Perform binarization processing on the grayscale channel data of the set-top box image data to obtain the binarized image data corresponding to the set-top box image data.

[0085] In an embodiment of the present application, the grayscale channel data of the set-top box image data can also be subjected to binarization processing to obtain the binarized image data corresponding to the set-top box image data. Exemplarily, the pixel values in the binarized image data can be either 0 and 1 or 0 and 255, which is not specifically limited and can be selected according to actual requirements.

[0086] Step S121c: Perform at least one morphological combination operation on the binarized image data to obtain at least one morphological image data corresponding to the binarized image data.

[0087] In an embodiment of the present application, after obtaining the binarized image data, at least one morphological combination operation can be performed on the binarized image data to obtain at least one morphological image data corresponding to the binarized image data. Each of the morphological combination operations includes one dilation operation and one erosion operation, and the size of the dilation kernel corresponding to the dilation operation can be larger than the size of the erosion kernel corresponding to the erosion operation. For example, the size of the dilation kernel can be 5*5, and the size of the erosion kernel can be 3*3. In addition, when performing multiple morphological combination operations, they can be performed separately, so as to avoid the problem that the image data is overly distorted when cascaded successively (that is, the object of the latter morphological combination operation is the result of the previous morphological combination operation). And, in order to ensure that more information can be obtained through multiple morphological combination operations, at least one of the sizes of the dilation kernel and the erosion kernel corresponding to every two morphological combination operations is different.

[0088] Step S121d: The convolution unit performs convolution processing on the grayscale map channel data to form a grayscale map convolution vector corresponding to the grayscale map channel data, and the convolution unit respectively performs convolution processing on each piece of the morphological image data to form a morphological image convolution vector corresponding to each piece of the morphological image data.

[0089] In the embodiment of the present application, the convolution unit can perform convolution processing on the grayscale map channel data to form a grayscale map convolution vector corresponding to the grayscale map channel data, and the convolution unit respectively performs convolution processing on each piece of the morphological image data to form a morphological image convolution vector corresponding to each piece of the morphological image data. It should be noted that since different image data represent different key information, therefore, when performing various convolution processes, it can be implemented by different convolution sub-units in the convolution unit, and, for the convenience of subsequent processing, the sizes of the vectors output by each convolution sub-unit can be the same.

[0090] Step S121e: Perform correlation mining processing on the grayscale map convolution vector corresponding to the grayscale map channel data and the morphological image convolution vector corresponding to each piece of the morphological image data, and output a channel data convolution vector corresponding to the grayscale map channel data.

[0091] In the embodiment of the present application, after obtaining the grayscale map convolution vector and the morphological image convolution vector, correlation mining processing can be performed on the grayscale map convolution vector corresponding to the grayscale map channel data and the morphological image convolution vector corresponding to each piece of the morphological image data, and a channel data convolution vector corresponding to the grayscale map channel data is output. That is to say, the information in the morphological image convolution vector corresponding to each piece of the morphological image data can be fused into the grayscale map convolution vector corresponding to the grayscale map channel data.

[0092] It can be understood that in the above step S121e, the specific manner of performing correlation mining processing on the grayscale map convolution vector corresponding to the grayscale map channel data and the morphological image convolution vector corresponding to each piece of the morphological image data is not limited, and can also be selected according to actual needs. For example, in an alternative embodiment, in order to enable the formed channel data convolution vector to have a high-accuracy semantic representation ability, the above step S121e can further include the following content:

[0093] First, for each morphological image convolution vector corresponding to the morphological image data, multiply the morphological image convolution vector by the restored image convolution vector corresponding to the grayscale image channel data to obtain a corresponding correlation parameter distribution, and based on this correlation parameter distribution, perform a weighted calculation on the restored image convolution vector corresponding to the grayscale image channel data, and output a morphological image correlation mining vector corresponding to the grayscale image channel data. The specific processing process is as described above;

[0094] Secondly, the mean value calculation or splicing process can be performed on each morphological image correlation mining vector corresponding to the grayscale image channel data and the grayscale image convolution vector corresponding to the grayscale image channel data to form a channel data convolution vector corresponding to the grayscale image channel data.

[0095] Regarding step S130, it should be noted that the specific manner of performing feature recognition processing on the set-top box image semantic vector in combination with the reference set-top box image data is not limited and can be selected according to actual needs.

[0096] For example, in an alternative embodiment, the reference set-top box image data can be subjected to feature extraction to obtain a corresponding semantic vector, and then, the vector distance between this semantic vector and the set-top box image semantic vector can be calculated. In this way, based on this vector distance, it can be determined whether it is qualified. If the vector distance is greater than the preset distance, it can be determined as unqualified; on the contrary, if the vector distance is less than or equal to the preset distance, it can be determined as qualified. In this way, the efficiency of feature recognition processing can be improved.

[0097] For another example, in another alternative embodiment, in order to ensure the reliability of the feature recognition processing, the above step S130 may further include step S131, step S132, step S133, and step S134. The specific content of each step is as described below.

[0098] Step S131: Through the feature extraction model, perform feature extraction processing on the reference set-top box image data, and output a reference image semantic vector corresponding to the reference set-top box image data.

[0099] In the embodiments of the present application, the reference set-top box image data can be subjected to feature extraction processing through the feature extraction model to output a reference image semantic vector corresponding to the reference set-top box image data. The specific processing process can refer to the processing process of the set-top box image data described above and will not be elaborated here one by one.

[0100] Step S132: Through the embedding unit included in the feature recognition model in the information recognition network, perform feature embedding processing on the set-top box description text corresponding to the qualified set-top box-related information in the reference set-top box image data, and output the description text embedding vector corresponding to the set-top box description text.

[0101] In an embodiment of the present application, it is also possible to perform feature embedding processing on the set-top box description text corresponding to the qualified set-top box-related information in the reference set-top box image data through the embedding unit included in the feature recognition model in the information recognition network, and output the description text embedding vector corresponding to the set-top box description text. For example, through a word embedding model, perform embedding processing on each word in the set-top box description text to obtain the corresponding word embedding vector. Then, the word embedding vectors can be subjected to mean calculation or splicing processing to obtain the corresponding description text embedding vector.

[0102] Step S133: Through the correlation mining unit included in the feature recognition model, based on the description text embedding vector, perform correlation mining on the reference image semantic vector, and output the reference image mining vector corresponding to the reference image semantic vector.

[0103] In an embodiment of the present application, after obtaining the description text embedding vector and the reference image semantic vector, through the correlation mining unit included in the feature recognition model, based on the description text embedding vector, perform correlation mining on the reference image semantic vector, and output the reference image mining vector corresponding to the reference image semantic vector. In this way, the semantic information in the description text embedding vector can be fused into the reference image semantic vector, thereby forming a reference image mining vector that fully represents the information content itself.

[0104] Step S134: Based on the vector distance between the reference image mining vector and the set-top box image semantic vector, determine the target set-top box quality data corresponding to the target IPTV set-top box, so as to complete the feature recognition processing of the set-top box image semantic vector.

[0105] In the embodiment of the present application, after obtaining the reference image mining vector, the target set-top box quality data corresponding to the target IPTV set-top box can be determined based on the vector distance (such as cosine distance, etc.) between the reference image mining vector and the set-top box image semantic vector, so as to complete the feature recognition process of the set-top box image semantic vector. For example, when the vector distance is greater than the preset distance, it can be determined as unqualified; when the vector distance is less than or equal to the preset distance, it can be determined as qualified. Additionally, in other embodiments, the reference image mining vector and the set-top box image semantic vector can also be concatenated, and then, convolution, pooling, and fully connected processing are performed. Finally, the results of the fully connected processing are classified and output to obtain the probabilities of qualified and unqualified, where the classification output can be implemented through functions such as softmax.

[0106] It can be understood that in the above step S133, the specific manner of performing correlation mining on the reference image semantic vector is not limited and can be selected accordingly according to actual needs. For example, in an alternative embodiment, in order to capture more detailed information through correlation mining, the above step S133 may further include step S133a and step S133b, and the specific content of each step is described as follows.

[0107] Step S133a: Through the correlation mining unit included in the feature recognition model, perform multiple-depth correlation mining on the reference image semantic vector based on the description text embedding vector, and output multiple depth correlation mining vectors corresponding to the reference image semantic vector.

[0108] In the embodiment of the present application, through the correlation mining unit included in the feature recognition model, multiple-depth correlation mining can be performed on the reference image semantic vector based on the description text embedding vector, and multiple depth correlation mining vectors corresponding to the reference image semantic vector can be output. In this way, semantic information at different depths can be captured, and the semantic representation ability can be improved.

[0109] Step S133b: Perform mean calculation or concatenation processing on the multiple depth correlation mining vectors corresponding to the reference image semantic vector, and output the reference image mining vector corresponding to the reference image semantic vector.

[0110] In the embodiment of the present application, after obtaining multiple depth correlation mining vectors, mean calculation or concatenation processing can be performed on the multiple depth correlation mining vectors corresponding to the reference image semantic vector, and the reference image mining vector corresponding to the reference image semantic vector can be output.

[0111] It can be understood that in the above step S133a, the specific manner of performing correlation mining on the reference image semantic vector at multiple depths is not limited and can also be selected according to actual needs. For example, in an alternative implementation manner, in order to capture more associated detailed information during the process of fusing the semantic information in the description text embedding vector into the reference image semantic vector, the above step S133a may further include the following content:

[0112] First, the description text embedding vector and the reference image semantic vector can be loaded into the correlation mining unit included in the feature recognition model. The specific processing process is as Figure 4 shown;

[0113] Second, in the correlation mining at the third depth, the description text embedding vector and the reference image semantic vector are multiplied to obtain the corresponding correlation parameter distribution. And based on this correlation parameter distribution, a weighted calculation is performed on the reference image semantic vector to output the first depth mining vector. The specific processing process is as described above;

[0114] Then, in the correlation mining at the second depth, the description text embedding vector and the first depth mining vector are respectively subjected to convolution processing to form the second depth convolutional description text embedding vector and the convolutional reference image semantic vector (the sizes after convolution processing can be the same). And the convolutional description text embedding vector and the convolutional reference image semantic vector are multiplied to obtain the corresponding correlation parameter distribution. And based on this correlation parameter distribution, a weighted calculation is performed on the convolutional reference image semantic vector to output the second depth mining vector;

[0115] After that, in the correlation mining at each depth after the second depth, the convolutional description text embedding vector and the depth mining vector at the previous depth are respectively subjected to convolution processing to form the convolutional description text embedding vector and the convolutional reference image semantic vector at the current depth. And the convolutional description text embedding vector and the convolutional reference image semantic vector are multiplied to obtain the corresponding correlation parameter distribution. And based on this correlation parameter distribution, a weighted calculation is performed on the convolutional reference image semantic vector to output the current depth mining vector;

[0116] Finally, deconvolution processing is respectively performed on the depth mining vectors at the second depth and each subsequent depth to obtain multiple vectors with the same size as the first depth mining vector. And these multiple vectors and the first depth mining vector are used as the multiple depth correlation mining vectors corresponding to the reference image semantic vector.

[0117] Finally, it should also be noted that in the above embodiments, when splicing the vectors, pooling processing can be performed on the vectors obtained by the splicing processing to achieve compression.

[0118] Combined with Figure 4 , the embodiment of the present application also provides an IPTV set-top box quality detection device applicable to the above electronic device. Among them, the IPTV set-top box quality detection device may include a set-top box image acquisition module, a feature extraction module, and a feature recognition module.

[0119] The set-top box image acquisition module is used to acquire the set-top box image data obtained by image acquisition of the target IPTV set-top box. Among them, the box body of the target IPTV set-top box has the set-top box related information formed by the printing process. In the embodiment of the present application, the set-top box image acquisition module can be used to execute Figure 2 the step S110 shown in

[0120] The feature extraction module is used to perform feature extraction processing on the set-top box image data through the feature extraction model in the information recognition network formed by pre-training, and output the set-top box image semantic vector corresponding to the set-top box image data. Among them, the information recognition network belongs to the neural network. In the embodiment of the present application, the feature extraction module can be used to execute Figure 2 the step S120 shown in

[0121] The feature recognition module is used to perform feature recognition processing on the set-top box image semantic vector through the feature recognition model in the information recognition network, combined with the reference set-top box image data, and output the target set-top box quality data corresponding to the target IPTV set-top box. Among them, the reference set-top box image data has qualified set-top box related information, and the target set-top box quality data is used to reflect whether the set-top box related information on the box body of the target IPTV set-top box is qualified. In the embodiment of the present application, the feature recognition module can be used to execute Figure 2 the step S130 shown in

[0122] In an embodiment of the present application, corresponding to the above IPTV set-top box quality detection method applied to the electronic device, a computer-readable storage medium is further provided. A computer program is stored in the computer-readable storage medium, and when the computer program runs, it executes each step of the IPTV set-top box quality detection method. Among them, the steps executed when the foregoing computer program runs will not be elaborated one by one here, and reference can be made to the foregoing explanation of the IPTV set-top box quality detection method.

[0123] In summary, for the IPTV set-top box quality detection method, device, equipment and medium provided by the present application, first, obtain the set-top box image data obtained by image acquisition of the target IPTV set-top box; secondly, perform feature extraction processing on the set-top box image data through the feature extraction model in the information recognition network formed by pre-training, and output the set-top box image semantic vector corresponding to the set-top box image data; then, through the feature recognition model in the information recognition network, combine the reference set-top box image data, perform feature recognition processing on the set-top box image semantic vector, and output the target set-top box quality data corresponding to the target IPTV set-top box. Based on the above content, on the one hand, the powerful learning ability of the neural network can be utilized to enable reliable feature recognition processing, so as to obtain reliable target set-top box quality data; on the other hand, since the reference set-top box image data is also used as an auxiliary during the feature recognition processing, the reliability of the feature recognition processing is improved. Therefore, it can effectively ensure that the target set-top box quality data can truly and reliably reflect the actual situation of the set-top box, thereby improving the relatively low reliability of set-top box quality detection in the existing technology.

[0124] In several embodiments provided by the embodiments of the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device and method embodiments described above are only illustrative. For example, the flowcharts and block diagrams in the drawings show the possible architectures, functions and operations of devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, program segment, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, as well as the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0125] In addition, each functional module in various embodiments of the present application may be integrated together to form an independent part, or each module may exist alone, or two or more modules may be integrated to form an independent part.

[0126] If the above-mentioned function is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, an electronic device, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs, etc., which can store program codes. It should be noted that in this article, the term "including", "comprising", or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such a process, method, article, or device. Without further limitations, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article, or device including the said element.

[0127] The above are only the preferred embodiments of the present application and are not used to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. A method for detecting quality of an IPTV set-top box, characterized in that: include: Acquire set-top box image data obtained by performing image acquisition on a target IPTV set-top box, wherein the box body of the target IPTV set-top box has set-top box related information formed by a printing process; Performing feature extraction processing on the set-top box image data through a feature extraction model in a pre-trained information recognition network, and outputting a set-top box image semantic vector corresponding to the set-top box image data, wherein the information recognition network is a neural network; Through the feature recognition model in the information recognition network, in combination with the reference set-top box image data, feature recognition processing is performed on the set-top box image semantic vector, and target set-top box quality data corresponding to the target IPTV set-top box is output, wherein the reference set-top box image data contains qualified set-top box related information, and the target set-top box quality data is used to reflect whether the set-top box related information on the box body of the target IPTV set-top box is qualified.

2. The IPTV set-top box quality detection method according to claim 1, characterized in that: The step of performing feature extraction processing on the set-top box image data by using the feature extraction model in the information recognition network formed by pre-training and outputting the corresponding set-top box image semantic vector comprises: Through the convolution unit in the feature extraction model of the information recognition network formed by pre-training, convolution processing is performed on multiple channel data of the set-top box image data respectively to form multiple channel data convolution vectors corresponding to the set-top box image data, wherein the multiple channel data at least include red channel data, green channel data, blue channel data and grayscale channel data; Multiplying the channel data convolution vector corresponding to the grayscale image channel data and the channel data convolution vector corresponding to the red channel data to obtain a corresponding correlation parameter distribution, and, based on the correlation parameter distribution, performing weighted calculation on the channel data convolution vector corresponding to the grayscale image channel data to form a corresponding first correlation mining vector; Multiplying the channel data convolution vector corresponding to the grayscale image channel data and the channel data convolution vector corresponding to the green channel data to obtain a corresponding correlation parameter distribution, and, based on the correlation parameter distribution, performing weighted calculation on the channel data convolution vector corresponding to the grayscale image channel data to form a corresponding second correlation mining vector; Multiplying the channel data convolution vector corresponding to the grayscale image channel data and the channel data convolution vector corresponding to the blue channel data to obtain a corresponding correlation parameter distribution, and, based on the correlation parameter distribution, performing weighted calculation on the channel data convolution vector corresponding to the grayscale image channel data to form a corresponding third correlation mining vector; The channel data convolution vector corresponding to the grayscale image channel data, the first correlation mining vector, the second correlation mining vector and the third correlation mining vector are averaged or concatenated to form a corresponding set-top box image semantic vector.

3. The IPTV set-top box quality detection method according to claim 2, characterized in that: The convolution unit in the feature extraction model of the information recognition network formed by pre-training performs convolution processing on the multiple channel data of the set-top box image data respectively to form the convolution vectors of the multiple channel data corresponding to the set-top box image data, including: The convolution unit in the feature extraction model of the information recognition network formed by pre-training performs convolution processing on the red channel data, the green channel data and the blue channel data of the set-top box image data respectively to form a channel data convolution vector corresponding to the red channel data, a channel data convolution vector corresponding to the green channel data and a channel data convolution vector corresponding to the blue channel data; Binarizing the grayscale channel data of the set-top box image data to obtain binary image data corresponding to the set-top box image data; Performing at least one morphological combination operation on the binary image data to obtain at least one morphological image data corresponding to the binary image data, wherein each of the morphological combination operations includes one dilation operation and one erosion operation; The grayscale image channel data is convolved by the convolution unit to form a grayscale image convolution vector corresponding to the grayscale image channel data, and each of the morphological image data is convolved by the convolution unit to form a morphological image convolution vector corresponding to each of the morphological image data; Correlation mining is performed on the grayscale image convolution vector corresponding to the grayscale image channel data and the morphological image convolution vector corresponding to each of the morphological image data, and the channel data convolution vector corresponding to the grayscale image channel data is output.

4. The IPTV set-top box quality detection method according to claim 3, characterized in that: The step of performing correlation mining processing on the grayscale image convolution vector corresponding to the grayscale image channel data and the morphological image convolution vector corresponding to each of the morphological image data, and outputting the channel data convolution vector corresponding to the grayscale image channel data comprises: For each morphological image convolution vector corresponding to the morphological image data, multiply the morphological image convolution vector and the restoration image convolution vector corresponding to the grayscale image channel data to obtain a corresponding correlation parameter distribution, and based on the correlation parameter distribution, perform weighted calculation on the restoration image convolution vector corresponding to the grayscale image channel data to output a morphological image correlation mining vector corresponding to the grayscale image channel data; Mean calculation or concatenation processing is performed on each morphological image correlation mining vector corresponding to the grayscale image channel data and the grayscale image convolution vector corresponding to the grayscale image channel data to form a channel data convolution vector corresponding to the grayscale image channel data.

5. The IPTV set-top box quality detection method according to any one of claims 1 to 4, characterized in that: The step of performing feature recognition processing on the set-top box image semantic vector by using the feature recognition model in the information recognition network in combination with reference set-top box image data, and outputting target set-top box quality data corresponding to the target IPTV set-top box comprises: Performing feature extraction processing on the reference set-top box image data through the feature extraction model, and outputting a reference image semantic vector corresponding to the reference set-top box image data; Performing feature embedding processing on the set-top box description text corresponding to the qualified set-top box related information in the reference set-top box image data through an embedding unit included in the feature recognition model in the information recognition network, and outputting a description text embedding vector corresponding to the set-top box description text; By using the correlation mining unit included in the feature recognition model, based on the description text embedding vector, the reference image semantic vector is subjected to correlation mining, and a reference image mining vector corresponding to the reference image semantic vector is output; Based on the vector distance between the reference image mining vector and the set-top box image semantic vector, the target set-top box quality data corresponding to the target IPTV set-top box is determined to complete feature recognition processing on the set-top box image semantic vector.

6. The IPTV set-top box quality detection method according to claim 5, characterized in that: The step of performing correlation mining on the reference image semantic vector based on the description text embedding vector by the correlation mining unit included in the feature recognition model and outputting a reference image mining vector corresponding to the reference image semantic vector comprises: Through the correlation mining unit included in the feature recognition model, based on the description text embedding vector, multiple depth correlation mining is performed on the reference image semantic vector, and multiple depth correlation mining vectors corresponding to the reference image semantic vector are output; The plurality of depth-related mining vectors corresponding to the reference image semantic vector are averaged or concatenated, and the reference image mining vector corresponding to the reference image semantic vector is output.

7. The IPTV set-top box quality detection method according to claim 6, characterized in that: The step of performing multiple depth correlation mining on the reference image semantic vector based on the description text embedding vector by the correlation mining unit included in the feature recognition model, and outputting multiple depth correlation mining vectors corresponding to the reference image semantic vector includes: Loading the description text embedding vector and the reference image semantic vector into a correlation mining unit included in the feature recognition model; In the first deep correlation mining, the description text embedding vector and the reference image semantic vector are multiplied to obtain a corresponding correlation parameter distribution, and the reference image semantic vector is weighted calculated based on the correlation parameter distribution to output a first deep mining vector; In the second depth correlation mining, the description text embedding vector and the first depth mining vector are respectively convolved to form a second depth convolution description text embedding vector and a convolution reference image semantic vector, and the convolution description text embedding vector and the convolution reference image semantic vector are multiplied to obtain a corresponding correlation parameter distribution, and the convolution reference image semantic vector is weighted calculated based on the correlation parameter distribution to output a second depth mining vector; In the correlation mining of each depth after the second depth, the convolution description text embedding vector and the depth mining vector of the previous depth are respectively convolved to form the convolution description text embedding vector and the convolution reference image semantic vector of the current depth, and the convolution description text embedding vector and the convolution reference image semantic vector are multiplied to obtain the corresponding correlation parameter distribution, and the convolution reference image semantic vector is weighted calculated based on the correlation parameter distribution to output the current depth mining vector; Deconvolution processing is performed on the depth mining vectors of the second depth and each subsequent depth to obtain multiple vectors of the same size as the first depth mining vector, and the multiple vectors and the first depth mining vector are used as multiple depth correlation mining vectors corresponding to the reference image semantic vector.

8. An IPTV set-top box quality detection device, characterized in that: include: A set-top box image acquisition module, used to acquire set-top box image data obtained by collecting images of a target IPTV set-top box, wherein the box body of the target IPTV set-top box has set-top box related information formed by a printing process; A feature extraction module, used to perform feature extraction processing on the set-top box image data through a feature extraction model in a pre-trained information recognition network, and output a set-top box image semantic vector corresponding to the set-top box image data, wherein the information recognition network belongs to a neural network; The feature recognition module is used to perform feature recognition processing on the set-top box image semantic vector through the feature recognition model in the information recognition network in combination with the reference set-top box image data, and output the target set-top box quality data corresponding to the target IPTV set-top box, wherein the reference set-top box image data contains qualified set-top box related information, and the target set-top box quality data is used to reflect whether the set-top box related information on the box body of the target IPTV set-top box is qualified.

9. An electronic device, characterized in that: include: Memory for storing computer programs; A processor connected to the memory is used to execute the computer program stored in the memory to implement the IPTV set-top box quality detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is run, the IPTV set-top box quality detection method according to any one of claims 1 to 7 is executed.

Images