A deep learning-based QR code detection method and device

Abstract

This application provides a deep learning-based QR code detection method and apparatus. The method obtains a first feature map through multi-scale feature extraction and feature fusion. Based on the first feature map, a segmentation map of a second preset scale for a first preset channel is obtained. The segmentation map is used to obtain the core region of the QR code to be tested according to its set confidence range. Based on the segmentation map, a first detection box of the QR code is obtained. The first detection box is the minimum area bounding rectangle of the contour in the segmentation map, and it is used to detect and identify the QR code to be tested. This application only needs to obtain the core region of the QR code to be tested through the segmentation map, that is, the process of the segmentation map classifying the core region and non-core region of the QR code, reducing the process required to obtain the QR code detection box, thereby reducing the time spent obtaining the QR code detection box and improving the efficiency of QR code detection and recognition.

Description

Technical Field

[0001] This application relates to the field of QR code detection, and in particular to a QR code detection method and apparatus based on deep learning. Background Technology

[0002] A QR code is a black-and-white graphic that uses specific geometric shapes arranged in a certain pattern on a plane to record data symbols. Common QR codes include Quick Response Code (QR code) and Data Matrix Code (DM code). The QR code includes three position detection patterns, while the DM code includes an edge-finding area.

[0003] In some QR code detection implementations, firstly, multiple anchor frames are generated based on the QR code to be tested, which includes both QR and DM QR codes. Then, the anchor frames are classified and their deviation information is regressed. Finally, based on the classified and regressed anchor frames, the detection frame of the QR code to be tested is determined, and the QR code is recognized based on the detection frame.

[0004] However, in scenarios involving the generation of QR code detection boxes, the efficiency of QR code detection and recognition is low due to the long time required for anchor box classification and position regression of anchor box deviation information in the anchor point mechanism. Summary of the Invention

[0005] To address the issue of low efficiency in QR code detection and recognition due to the time-consuming process of classifying anchor frames and regressing their position information in the anchor point mechanism, this application provides a deep learning-based QR code detection method and apparatus.

[0006] To address the aforementioned technical problems, the embodiments of this application disclose the following technical solutions:

[0007] In a first aspect, embodiments of this application disclose a QR code detection method based on deep learning, comprising: performing multi-scale feature extraction on the QR code to be tested, obtaining a multi-scale feature map, using the multi-scale feature map to fuse and generate a first feature map, the first feature map being a high-dimensional feature map of a first preset scale, and the QR code to be tested including a QR code and a DM code;

[0008] Based on the first feature map, a segmentation map of the second preset scale of the first preset channel is obtained. The segmentation map is used to obtain the core area of ​​the QR code to be tested according to its set confidence range.

[0009] The segmentation image is subjected to channel-by-channel contour detection and contour analysis to obtain a first detection box, which includes DM QR codes or QR codes. The first detection box is the smallest bounding rectangle of the contour in the segmentation image. The first detection box is used to detect and identify the QR code to be tested.

[0010] Optionally, the segmentation image is subjected to channel-by-channel contour detection and contour analysis to obtain the first detection box of the DM QR code and the first detection box of the QR code, and the first detection box is displayed according to the type of the QR code to be tested.

[0011] Optionally, based on the first feature map, a threshold map of the second preset scale of the first preset channel is obtained. The threshold map is used to obtain the boundary region of the QR code to be tested according to its set confidence range.

[0012] Based on the segmentation map and the threshold map, obtain the soft binarization map;

[0013] The soft binarized image is subjected to channel-by-channel contour detection and contour analysis to obtain a second detection box, which includes DM QR codes or QR codes. The second detection box is the minimum area bounding rectangle of the contour in the soft binarized image. The second detection box is used to detect and identify the QR code to be tested.

[0014] Optionally, in the step of obtaining a soft binarized map based on the segmentation map and the threshold map, the method includes:

[0015] The segmentation map is subtracted from the threshold map channel by channel and element by element to obtain the residual map. The sigmoid operation is then performed on the residual map to obtain the soft binarization map.

[0016] Optionally, in the step of using multi-scale feature maps to fuse and generate a first feature map, wherein the first feature map is a high-dimensional feature map of a first preset scale, the method further includes:

[0017] Upsample the multi-scale feature maps to obtain multiple feature maps of the first preset scale. Based on the multiple feature maps of the first preset scale and the convolutional layer, obtain multiple feature maps of the first preset scale of the second preset channels. Fuse the feature maps of the first preset scale of the multiple second preset channels into a first feature map. The first feature map is a high-dimensional feature map of the first preset scale, and the number of channels in the first feature map is more than that in the second preset channels.

[0018] Secondly, this application discloses a deep learning-based QR code detection device, including: a feature extraction module, used to perform: multi-scale feature extraction on the QR code to be tested, to obtain a multi-scale feature map, the multi-scale feature map is used to fuse and generate a first feature map, the first feature map is a high-dimensional feature map of a first preset scale, and the QR code to be tested includes QR code and DM code;

[0019] The segmentation module is used to perform the following: based on the first feature map, obtain a segmentation map of the second preset scale for the first preset channel, and the segmentation map is used to obtain the core region of the QR code to be tested according to its set confidence range;

[0020] The first detection box generation module is used to perform: channel-by-channel contour detection and contour analysis on the segmentation image to obtain a first detection box including DM QR code or QR code. The first detection box is the smallest bounding rectangle of the contour in the segmentation image. The first detection box is used to detect and identify the QR code to be tested.

[0021] Optionally, the first detection box generation module is also used to perform:

[0022] Perform channel-by-channel contour detection and contour analysis on the segmented image to obtain the first detection box of the DM QR code and the first detection box of the QR code. Display the first detection box according to the type of the QR code to be tested.

[0023] Optionally, the device also includes:

[0024] The threshold module is used to perform the following: based on the first feature map, obtain a threshold map of the second preset scale of the first preset channel, and the threshold map is used to obtain the boundary region of the QR code to be tested according to its set confidence range;

[0025] The binarization module is used to perform the following: obtain a soft binarized image based on the segmentation image and the threshold image;

[0026] The second detection box generation module is used to perform: channel-by-channel contour detection and contour analysis on the soft binarized image to obtain a second detection box including DM QR codes or QR codes. The second detection box is the minimum area bounding rectangle of the contour in the soft binarized image. The second detection box is used to detect and identify the QR code to be tested.

[0027] Optional, a binarization module, specifically used for execution:

[0028] The segmentation map is subtracted from the threshold map channel by channel and element by element to obtain the residual map. The sigmoid operation is then performed on the residual map to obtain the soft binarization map.

[0029] Optional, feature extraction module, specifically used for execution:

[0030] Upsample the multi-scale feature maps to obtain multiple feature maps of the first preset scale. Based on the multiple feature maps of the first preset scale and the convolutional layer, obtain multiple feature maps of the first preset scale of the second preset channels. Fuse the feature maps of the first preset scale of the multiple second preset channels into a first feature map. The first feature map is a high-dimensional feature map of the first preset scale, and the number of channels in the first feature map is more than that in the second preset channels.

[0031] The beneficial effects of this application are as follows: This application obtains a first feature map through multi-scale feature extraction and feature fusion. Based on the first feature map, a segmentation map of a second preset scale for a first preset channel is obtained. The segmentation map is used to obtain the core region of the QR code to be tested according to its set confidence range. Based on the segmentation map, a first detection box of the QR code is obtained. The first detection box is the minimum area bounding rectangle of the contour in the segmentation map. The first detection box is used to detect and identify the QR code to be tested. This application only needs to obtain the core region of the QR code to be tested through the segmentation map. That is, the process of the segmentation map classifying the core region and non-core region of the QR code is compared with classifying anchor boxes sequentially and performing position regression on the deviation information of the anchor boxes. This reduces the process required to obtain the QR code detection box, thereby reducing the time spent obtaining the QR code detection box and improving the efficiency of QR code detection and recognition.

[0032] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0033] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 A flowchart illustrating a deep learning-based QR code detection method provided in this application embodiment;

[0035] Figure 2 A schematic diagram of a deep learning-based QR code detection device provided in an embodiment of this application;

[0036] Figure 3 This is a schematic diagram of another deep learning-based QR code detection device provided in an embodiment of this application. Detailed Implementation

[0037] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of this application.

[0038] To facilitate the explanation of the technical solution of this application, some concepts involved in this application will be explained first below.

[0039] The confidence level is the probability that the population parameter value falls within a certain range of the sample statistics. In this application, the confidence level is the probability that the core area of ​​the QR code to be tested falls within the segmentation map, or the boundary area of ​​the QR code to be tested falls within the threshold map.

[0040] See Figure 1 This application provides a deep learning-based QR code detection method, including the following steps:

[0041] Step S110: Perform multi-scale feature extraction on the QR code to be tested to obtain a multi-scale feature map. The multi-scale feature map is used to fuse and generate a first feature map. The first feature map is a high-dimensional feature map of a first preset scale. The QR code to be tested includes QR code and DM code.

[0042] In some embodiments, multi-scale feature maps are upsampled to obtain multiple feature maps at a first preset scale. Based on the multiple feature maps at the first preset scale and a convolutional layer, multiple feature maps at the first preset scale with second preset channels are obtained. The feature maps at the first preset scale with the multiple second preset channels are fused into a first feature map. The first feature map is a high-dimensional feature map at the first preset scale, and the number of channels in the first feature map is greater than that of the second preset channels. Because multi-scale feature extraction is performed, the detection accuracy of the QR code to be tested is improved.

[0043] In some embodiments, the QR code to be tested is subjected to multi-scale feature extraction through a mainstream backbone network. The mainstream backbone network includes the ResNet (Residual Neural Network) series network and the MobileNet series network. Both the ResNet series network and the MobileNet series network are network structures used for feature extraction. The multi-scale can be selected as 1 / 4 scale, 1 / 8 scale, 1 / 16 scale and 1 / 32 scale.

[0044] In some embodiments, upsampling of multi-scale feature maps to obtain multiple feature maps at first preset scales, wherein the multi-scale includes 1 / 4 scale, 1 / 8 scale, 1 / 16 scale, and 1 / 32 scale, includes:

[0045] The feature map at a scale of 1 / 32 extracted is upsampled to a feature map at a scale of 1 / 16. The upsampled feature map at a scale of 1 / 16 is then added pixel by pixel to the feature map at a scale of 1 / 16 extracted. Finally, convolution smoothing is performed on the feature map at a scale of 1 / 16 after pixel-by-pixel addition, resulting in a feature map at a scale of 1 / 16 after upsampling, element-by-element addition, and convolution smoothing.

[0046] The feature map at a scale of 1 / 16 extracted from the feature extraction is upsampled to a feature map at a scale of 1 / 8. The upsampled feature map at a scale of 1 / 8 is then added pixel by pixel to the feature map at a scale of 1 / 8 extracted from the feature extraction. Finally, a convolutional smoothing process is performed on the feature map at a scale of 1 / 8 after pixel-by-pixel addition, resulting in a feature map at a scale of 1 / 8 after upsampling, element-by-element addition, and convolutional smoothing.

[0047] The feature map at a scale of 1 / 8 extracted is upsampled to a feature map at a scale of 1 / 4. The upsampled feature map at a scale of 1 / 4 is then added pixel by pixel to the feature map at a scale of 1 / 4 extracted. Finally, a convolutional smoothing process is performed on the feature map at a scale of 1 / 4 after pixel-by-pixel addition, resulting in a feature map at a scale of 1 / 4 after upsampling, element-by-element addition, and convolutional smoothing.

[0048] Based on the feature map at a scale of 1 / 32 extracted from the feature extraction, and the feature maps at scales of 1 / 16, 1 / 8, and 1 / 4 after upsampling, element-wise addition, and convolutional smoothing, multiple feature maps at a first preset scale are obtained. The first preset scale can be selected as the 1 / 4 scale.

[0049] In some embodiments, the multi-scale includes a 1 / 4 scale, a 1 / 8 scale, a 1 / 16 scale, and a 1 / 32 scale. When the first preset scale is a 1 / 4 scale, the feature map at the 1 / 32 scale extracted by the feature extraction is upsampled by 8 times, the feature map at the 1 / 16 scale after upsampling, element-wise addition, and convolutional smoothing is upsampled by 4 times, and the feature map at the 1 / 8 scale after upsampling, element-wise addition, and convolutional smoothing is upsampled by 2 times. The feature map at the 1 / 4 scale after upsampling, element-wise addition, and convolutional smoothing remains unchanged, thus obtaining four 1 / 4 scale feature maps.

[0050] In some embodiments, feature maps of multiple second preset channels at the first preset scale are obtained based on feature maps of multiple first preset scales and convolutional layers. The first preset scale can be selected as 1 / 4 scale, and the second preset channels can be selected as C / 4 channels, where C is a positive integer greater than or equal to 4.

[0051] In some embodiments, when the plurality of first preset scale feature maps are four 1 / 4 scale feature maps and the second preset channel is a C / 4 channel, the four 1 / 4 scale feature maps are transformed into four C / 4 channel 1 / 4 scale feature maps through a convolutional layer; in the channel dimension, the four C / 4 channel 1 / 4 scale feature maps are stitched together to form a 1 / 4 scale C channel first feature map.

[0052] Step S120: Based on the first feature map, obtain a segmentation map of the second preset scale of the first preset channel. The segmentation map is used to obtain the core area of ​​the QR code to be tested according to its set confidence range.

[0053] In some embodiments, a segmentation map of the second preset scale of the first preset channel is obtained based on the first feature map and the transposed convolution, and the segmentation map can be selected as a dual-channel 1 / 1 scale segmentation map.

[0054] In some embodiments, the confidence range is the pixel value range. When the pixel values ​​of all pixels in the segmentation image are 0-1, the pixels with pixel values ​​in the segmentation image within the range of 0.5-1 are obtained, and the region where the pixel is located is obtained as the core region of the QR code to be tested.

[0055] In some embodiments, the segmentation map, similar to a mask, is used to distinguish regions with a set confidence range from regions without a set confidence range.

[0056] Step S130: Perform contour detection and contour analysis on the segmentation image channel by channel to obtain a first detection box including DM QR code or QR code. The first detection box is the minimum area bounding rectangle of the contour in the segmentation image. The first detection box is used to detect and identify the QR code to be tested.

[0057] In some embodiments, contour detection and contour analysis are performed on the segmentation map channel by channel to obtain the first detection box of the DM QR code and the first detection box of the QR code, and the first detection box is displayed according to the type of the QR code to be tested.

[0058] In some embodiments, a threshold map of a second preset scale for a first preset channel is obtained based on a first feature map. The threshold map is used to obtain the boundary region of the QR code to be tested according to its set confidence range.

[0059] Based on the segmentation map and the threshold map, obtain the soft binarization map;

[0060] The soft binarized image is subjected to channel-by-channel contour detection and contour analysis to obtain a second detection box, which includes DM QR codes or QR codes. The second detection box is the minimum area bounding rectangle of the contour in the soft binarized image. The second detection box is used to detect and identify the QR code to be tested.

[0061] Since the binarization threshold (i.e., the threshold map) is learned and adaptive, heuristic methods such as pixel clustering before generating the detection box of the QR code can be eliminated, reducing the process required to obtain the QR code detection box, thereby reducing the time spent obtaining the QR code detection box and improving the efficiency of QR code detection and recognition.

[0062] In some embodiments, the confidence range is the pixel value range. When the pixel values ​​of all pixels in the threshold image are 0-1, the pixels with pixel values ​​in the threshold image within the range of 0.5-1 are obtained, and the region where the pixel is located is obtained as the boundary region of the QR code to be tested.

[0063] In some embodiments, the threshold map, similar to a mask, is used to distinguish regions with a set confidence range from regions without a set confidence range.

[0064] In some embodiments, the segmentation map is subtracted from the threshold map channel by channel and element by element to obtain a residual map. A sigmoid operation is then performed on the residual map to obtain a soft binarized map. In the sigmoid operation, the sigmoid function curve can be shrunk according to actual needs, i.e., sigmoid = 1 / (1+e^(-kx)). When k = 1, it is a standard sigmoid function. The shrinkage is achieved by adjusting k.

[0065] like Figure 2 As shown, corresponding to the aforementioned embodiment of a deep learning-based QR code detection method, this application also provides an embodiment of a deep learning-based QR code detection device. The device includes:

[0066] Feature extraction module 1 is used to perform: multi-scale feature extraction on the QR code to be tested, obtain multi-scale feature maps, and use the multi-scale feature maps to fuse and generate a first feature map. The first feature map is a high-dimensional feature map of a first preset scale. The QR code to be tested includes QR codes and DM codes.

[0067] In some embodiments, feature extraction module 1 is specifically used to perform:

[0068] Upsample the multi-scale feature maps to obtain multiple feature maps of the first preset scale. Based on the multiple feature maps of the first preset scale and the convolutional layer, obtain multiple feature maps of the first preset scale of the second preset channels. Fuse the feature maps of the first preset scale of the multiple second preset channels into a first feature map. The first feature map is a high-dimensional feature map of the first preset scale, and the number of channels in the first feature map is more than that in the second preset channels.

[0069] The segmentation module 2 is used to perform the following: based on the first feature map, obtain a segmentation map of the second preset scale of the first preset channel. The segmentation map is used to obtain the core area of ​​the QR code to be tested according to its set confidence range.

[0070] The first detection box generation module 3 is used to perform: channel-by-channel contour detection and contour analysis on the segmentation image to obtain a first detection box including DM QR code or QR code. The first detection box is the smallest bounding rectangle of the contour in the segmentation image. The first detection box is used to detect and identify the QR code to be tested.

[0071] In some embodiments, the first detection box generation module 3 is further configured to perform:

[0072] Perform channel-by-channel contour detection and contour analysis on the segmented image to obtain the first detection box of the DM QR code and the first detection box of the QR code. Display the first detection box according to the type of the QR code to be tested.

[0073] In some embodiments, such as Figure 3 As shown, the device also includes:

[0074] Threshold module 4 is used to perform the following: based on the first feature map, obtain a threshold map of the second preset scale of the first preset channel, and the threshold map is used to obtain the boundary region of the QR code to be tested according to its set confidence range;

[0075] Binarization module 5 is used to perform the following: obtain a soft binarized image based on the segmentation image and the threshold image;

[0076] The second detection box generation module 6 is used to perform: channel-by-channel contour detection and contour analysis on the soft binarized image to obtain a second detection box including DM QR code or QR code. The second detection box is the minimum area bounding rectangle of the contour in the soft binarized image. The second detection box is used to detect and identify the QR code to be tested.

[0077] In some embodiments, binarization module 5 is specifically used to perform:

[0078] The segmentation image is subtracted from the threshold image channel by channel and element by element to obtain the residual image. A sigmoid operation is then performed on the residual image to obtain a soft binarization image. Since the binarization threshold (i.e., the threshold image) is learned and adaptive, heuristic methods such as pixel clustering before generating the QR code detection box can be omitted, reducing the process required to obtain the QR code detection box, thereby reducing the time spent obtaining the QR code detection box and improving the efficiency of QR code detection and recognition.

[0079] Since the above embodiments are all described in conjunction with other methods, and different embodiments have the same parts, the same or similar parts between the various embodiments in this specification can be referred to mutually. They will not be described in detail here.

[0080] It should be noted that, in this specification, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a circuit structure, article, or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a circuit structure, article, or device. Without further limitations, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the circuit structure, article, or device that includes said element.

[0081] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the content of the claims.

[0082] The embodiments described above do not constitute a limitation on the scope of protection of this application.

Claims

1. A deep learning-based QR code detection method, characterized in that, include: Multi-scale feature extraction is performed on the QR code to be tested to obtain a multi-scale feature map. The multi-scale feature map is used to fuse and generate a first feature map. The first feature map is a high-dimensional feature map of a first preset scale. The QR code to be tested includes QR code and DM code. Based on the first feature map, a segmentation map of the second preset scale of the first preset channel is obtained. The segmentation map is used to obtain the core area of ​​the QR code to be tested according to its set confidence range. The segmentation image is subjected to channel-by-channel contour detection and contour analysis to obtain a first detection box including a DM QR code or a QR code. The first detection box is the minimum area bounding rectangle of the contour in the segmentation image. The first detection box is used to detect and identify the QR code to be tested. The method further includes: Based on the first feature map, a threshold map of the second preset scale of the first preset channel is obtained. The threshold map is used to obtain the boundary region of the QR code to be tested according to its set confidence range. Based on the segmentation map and the threshold map, a soft binarization map is obtained; The soft binarized image is subjected to channel-by-channel contour detection and contour analysis to obtain a second detection box including a DM QR code or a QR code. The second detection box is the minimum area bounding rectangle of the contour in the soft binarized image. The second detection box is used to detect and identify the QR code to be tested. In the step of obtaining a soft binarized map based on the segmentation map and the threshold map, the method includes: The segmentation map is subtracted from the threshold map channel by channel and element by element to obtain the residual map. The residual map is then subjected to a sigmoid operation to obtain a soft binarization map.

2. The deep learning-based QR code detection method according to claim 1, characterized in that, The method includes: The segmented image is subjected to channel-by-channel contour detection and contour analysis to obtain the first detection box of the DM QR code and the first detection box of the QR code. The first detection box is displayed according to the type of the QR code to be tested.

3. The deep learning-based QR code detection method according to claim 1, characterized in that, In the step of fusing the multi-scale feature maps to generate a first feature map, wherein the first feature map is a high-dimensional feature map of a first preset scale, the method further includes: Upsampling of the multi-scale feature map yields multiple feature maps at a first preset scale. Based on the multiple feature maps at the first preset scale and the convolutional layer, multiple feature maps at the first preset scale with second preset channels are obtained. The feature maps at the first preset scale with the multiple second preset channels are fused into the first feature map. The first feature map is a high-dimensional feature map at the first preset scale, and the number of channels in the first feature map is greater than that of the second preset channels.

4. A QR code detection device based on deep learning, characterized in that, include: The feature extraction module is used to perform: multi-scale feature extraction on the QR code to be tested to obtain a multi-scale feature map, the multi-scale feature map is used to fuse and generate a first feature map, the first feature map is a high-dimensional feature map of a first preset scale, and the QR code to be tested includes QR code and DM code; The segmentation module is used to perform the following: based on the first feature map, obtain a segmentation map of the second preset scale for the first preset channel, wherein the segmentation map is used to obtain the core region of the QR code to be tested according to its set confidence range; The first detection box generation module is used to perform: channel-by-channel contour detection and contour analysis on the segmentation image to obtain a first detection box including a DM QR code or a QR code. The first detection box is the minimum area bounding rectangle of the contour in the segmentation image. The first detection box is used to detect and identify the QR code to be tested. The device further includes: The threshold module is used to perform the following: based on the first feature map, obtain a threshold map of the second preset scale of the first preset channel, wherein the threshold map is used to obtain the boundary region of the QR code to be tested according to its set confidence range; The binarization module is used to perform the following: obtain a soft binarization map based on the segmentation map and the threshold map; The second detection box generation module is used to perform: channel-by-channel contour detection and contour analysis on the soft binarized image to obtain a second detection box including a DM QR code or a QR code. The second detection box is the minimum area bounding rectangle of the contour in the soft binarized image. The second detection box is used to detect and identify the QR code to be tested. The binarization module is specifically used to perform: The segmentation map is subtracted from the threshold map channel by channel and element by element to obtain the residual map. The residual map is then subjected to a sigmoid operation to obtain a soft binarization map.

5. The deep learning-based QR code detection device according to claim 4, characterized in that, The first detection box generation module is also used to perform: The segmented image is subjected to channel-by-channel contour detection and contour analysis to obtain the first detection box of the DM QR code and the first detection box of the QR code. The first detection box is displayed according to the type of the QR code to be tested.

6. The deep learning-based QR code detection device according to claim 4, characterized in that, The feature extraction module is specifically used to perform: Upsampling of the multi-scale feature map yields multiple feature maps at a first preset scale. Based on the multiple feature maps at the first preset scale and the convolutional layer, multiple feature maps at the first preset scale with second preset channels are obtained. The feature maps at the first preset scale with the multiple second preset channels are fused into the first feature map. The first feature map is a high-dimensional feature map at the first preset scale, and the number of channels in the first feature map is greater than that of the second preset channels.

Images