CN-121998978-A - Correction prompt-based method, device and storage medium for photographing and warehousing along with manifest

CN121998978ACN 121998978 ACN121998978 ACN 121998978ACN-121998978-A

Abstract

The invention discloses a correction prompt-based on-demand bill shooting and warehousing method, a correction prompt-based on-demand bill shooting and warehousing device and a storage medium, wherein the quality evaluation of target on-demand bill images is realized by utilizing an image quality evaluation model, and when the quality does not accord with preset conditions, image shooting correction guiding prompt information is output to prompt a user to carry out correction shooting, so that shooting correction guiding functions are realized; meanwhile, after the acquired image quality of the target along with the manifest image accords with a preset condition, extracting a differential image used for representing a difference part between the target along with the manifest image and the reference along with the manifest image, uploading the differential image only, and finally restoring the target along with the manifest image by utilizing the differential image and the reference along with the manifest image at a cloud end to finish warehousing of the manifest.

Inventors

JIANG XIAOFENG
MO SHA
CHEN YI
LI DONGGEN

Assignees

成都字节流科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260409

Claims (10)

1. The method for photographing and warehousing along with the manifest based on correction prompt is characterized by comprising the following steps: acquiring a target manifest-following image; Adopting an image quality evaluation model to evaluate the image quality of the target along with the manifest image to obtain an image quality evaluation result; Judging whether the image quality of the target manifest-following image meets a preset condition or not based on an image quality evaluation result; If yes, acquiring a reference along-with-bill image, and extracting a difference image between the target along-with-bill image and the reference along-with-bill image, otherwise, outputting image shooting correction guide prompt information to reacquire the target along-with-bill image after outputting the image shooting correction guide prompt information until the image quality of the target along-with-bill image meets the preset condition, wherein the difference image is used for representing an image corresponding to a difference part between the target along-with-bill image and the reference along-with-bill image; compressing the differential image to obtain a compressed image; Transmitting the compressed image to a cloud end so that the cloud end can reconstruct the target manifest-following image based on the compressed image and the reference manifest-following image, and storing the target manifest-following image into a database.
2. The method of claim 1, wherein the image quality assessment model comprises an initial convolution layer, an inverse residual bottleneck block layer, a global averaging pooling layer, and a full connection layer, wherein the full connection layer comprises a first branch connection layer, a second branch connection layer, and a third branch connection layer, and each branch connection layer is provided with a respective corresponding activation function; The initial convolution layer is used for carrying out feature extraction processing on the input target manifest-following image to obtain a first feature image; The reverse residual bottleneck block layer is used for carrying out feature re-extraction processing on the first feature image based on a depth separable convolution mechanism and an attention mechanism to obtain a second feature image; the global average pooling layer is used for carrying out global average pooling treatment on the second characteristic image to obtain one-dimensional characteristic vectors, and inputting the one-dimensional characteristic vectors into the three branch connecting layers respectively; The first branch connection layer is used for mapping the one-dimensional feature vector by utilizing a corresponding activation function to obtain the definition score of the target manifest-following image; the second branch connecting layer is used for mapping the one-dimensional feature vector by utilizing a corresponding activation function to obtain a sine value and a cosine value of a yaw angle of the target manifest-following image; The third branch connection layer is used for mapping the one-dimensional feature vector by utilizing a corresponding activation function to obtain the illumination level of the target manifest-following image; And the image quality evaluation result of the target manifest image is formed by utilizing the definition score, the illumination level and the sine value and the cosine value of the yaw angle of the target manifest image.
3. The method according to claim 2, wherein the residual pouring bottleneck block layer comprises a bottleneck block and a plurality of MBConv blocks connected in sequence, the output feature of any MBConv block is subjected to feature fusion with the input feature of any MBConv block, the obtained fusion feature is output to the next MBConv block, the fusion feature output by the last MBConv block is input to the bottleneck block, and the bottleneck block is used for outputting the second feature image; Wherein, any MBConv blocks comprise a1×1 updimension convolution layer, a3×3 first depth separable convolution layer, a 5×5 second depth separable convolution layer, a SE attention layer, and a1×1 dimension reduction convolution layer, which are connected in sequence.
4. The method of claim 2, wherein the initial convolutional layer is a 3x3 two-dimensional convolutional layer, wherein the two-dimensional convolutional layer has a convolution step size of 3, the number of convolution kernels is 16, and the inverse residual bottleneck block layer comprises 8 MBConv blocks.
5. The method of claim 2, wherein determining whether the image quality of the target manifest-in-manifest image meets a preset condition based on the image quality evaluation result comprises: Calculating an angle deviation value of the target manifest-following image based on the sine value and the cosine value of the yaw angle; Normalizing the definition score, the angle deviation value and the illumination level to obtain normalized definition, normalized angle deviation and normalized illumination level; Carrying out weighted summation processing on the normalized definition, the normalized angle deviation and the normalized illumination level to obtain a quality score of the target manifest-following image; Judging whether the quality score is larger than or equal to a quality threshold value; if yes, judging that the image quality of the target manifest-following image meets a preset condition.
6. The method of claim 5, wherein the quality threshold is determined by; carrying out manifest type identification on the target manifest image to obtain the manifest type of the target manifest image, wherein the manifest type comprises a standard printing manifest, a handwriting manifest and a damaged manifest; obtaining a threshold mapping dictionary, wherein score thresholds corresponding to different manifest types are stored in the threshold mapping dictionary; and matching a score threshold corresponding to the manifest type in the threshold mapping dictionary according to the manifest type, and taking the matched score threshold as the quality threshold.
7. The method of claim 1, wherein extracting a difference image between the target manifest image and the reference manifest image comprises: Performing feature point matching processing on the target along with the manifest image and the reference along with the manifest image to obtain matching feature pairs; According to the matching feature pairs, calculating a perspective transformation matrix, and based on the perspective transformation matrix, aligning the target manifest-following image with the reference manifest-following image to obtain an aligned target manifest-following image; Performing differential operation on the aligned target along with the manifest image and the reference along with the manifest image to obtain an initial differential image; Performing contour extraction on the initial differential image to obtain a differential contour image, and taking the differential contour image as the differential image; The compressed image is transmitted to the cloud end, and then includes: Acquiring ID information of a reference manifest-following image; generating a data packet by using the compressed image, the perspective transformation matrix and the ID information of the reference manifest-following image; Transmitting the data packet to a cloud end, so that the cloud end calls the reference manifest image based on ID information in the data packet, decompresses the compressed image to obtain a differential image, and pastes the differential image into the reference manifest image to obtain a reconstructed image, so that the reconstructed image is subjected to image correction based on a perspective transformation matrix to restore the target manifest image.
8. The utility model provides a warehouse entry device of shooing along with manifest based on correction suggestion which characterized in that includes: The acquisition unit is used for acquiring the target manifest-following image; the image quality evaluation unit is used for carrying out image quality evaluation on the target along with the manifest image by adopting an image quality evaluation model to obtain an image quality evaluation result; The image quality evaluation unit is further used for judging whether the image quality of the target manifest-following image meets a preset condition or not based on an image quality evaluation result; the image processing unit is used for acquiring a reference follow-up bill image when the image quality evaluation unit determines that the image quality meets the preset condition, extracting a difference image between the target follow-up bill image and the reference follow-up bill image, and outputting image shooting correction guide prompt information when the image quality evaluation unit determines that the image quality does not meet the preset condition, so as to re-acquire the target follow-up bill image after the image shooting correction guide prompt information is output until the image quality of the target follow-up bill image meets the preset condition, wherein the difference image is used for representing an image corresponding to a difference part between the target follow-up bill image and the reference follow-up bill image; The compression unit is used for carrying out compression processing on the differential image to obtain a compressed image; the transmission unit is used for transmitting the compressed image to the cloud end so that the cloud end can reconstruct the target manifest-following image based on the compressed image and the reference manifest-following image, and the target manifest-following image is stored in the database.
9. The correction prompt-based on-demand bill photographing and warehousing device is characterized by comprising a memory, a processor and a transceiver which are sequentially connected in a communication mode, wherein the memory is used for storing a computer program, the transceiver is used for receiving and transmitting messages, and the processor is used for reading the computer program and executing the correction prompt-based on-demand bill photographing and warehousing method according to any one of claims 1-7.
10. A storage medium having instructions stored thereon that, when executed on a computer, perform the corrective prompt-based on-demand manifest shooting warehousing method of any one of claims 1-7.

Description

Correction prompt-based method, device and storage medium for photographing and warehousing along with manifest Technical Field The invention belongs to the technical field of image recognition, and particularly relates to a correction prompt-based method, a correction prompt-based device and a storage medium for photographing and warehousing along with a manifest. Background According to the 2024 report of the domestic logistics and purchasing union, three technical routes mainly exist for the current in-industry manifest identification, namely (1) a manual input mode, which takes up 37.2% of the total weight of the system, the average processing time is 28 seconds/manifest, the error rate is 6.8% -12.4% (related to the proficiency of operators), and (2) cloud OCR (optical character recognition), which takes up 52.1% of the total weight of the system, are typical systems, namely WMS 3.0 of a certain logistics company and the warehouse brain of a certain e-commerce group, are sensitive to network delay, and when the network delay is more than 150ms, the identification failure rate is increased to 24.7%, and (3) localized identification equipment, which takes up 10.7% of the total weight of the system, and mainly has the problems of high cost and poor flexibility (the unit price of special equipment is more than 8000 yuan and can not adapt to manifest of different specifications). The method has the following defects that (1) the real-time performance is insufficient, the feedback can be obtained only after the complete image is uploaded to the cloud end by the existing mobile terminal scheme, the average response time is 6.3 seconds (under the 4G network environment), (2) the quality evaluation is lack, 78% of recognition errors are caused by the image quality problem (blurring, tilting and overexposure), but the existing system has no shooting guidance function, (3) the resource waste is serious, and the test shows that 42% of uploaded image contains more than 60% of invalid background area, so that how to provide the real-time performance is high, the resource can be saved, the quality evaluation can be carried out, and the bill-following shooting warehousing method with the shooting guidance function becomes a problem to be solved urgently based on the defects. Disclosure of Invention The invention aims to provide a correction prompt-based method, device and storage medium for photographing and warehousing along with a manifest, which are used for solving the problems of insufficient real-time performance, quality assessment deficiency and serious resource waste in the prior art. In order to achieve the above purpose, the present invention adopts the following technical scheme: In a first aspect, a method for photographing and warehousing along with a manifest based on correction prompt is provided, including: acquiring a target manifest-following image; Adopting an image quality evaluation model to evaluate the image quality of the target along with the manifest image to obtain an image quality evaluation result; Judging whether the image quality of the target manifest-following image meets a preset condition or not based on an image quality evaluation result; If yes, acquiring a reference along-with-bill image, and extracting a difference image between the target along-with-bill image and the reference along-with-bill image, otherwise, outputting image shooting correction guide prompt information to reacquire the target along-with-bill image after outputting the image shooting correction guide prompt information until the image quality of the target along-with-bill image meets the preset condition, wherein the difference image is used for representing an image corresponding to a difference part between the target along-with-bill image and the reference along-with-bill image; compressing the differential image to obtain a compressed image; Transmitting the compressed image to a cloud end so that the cloud end can reconstruct the target manifest-following image based on the compressed image and the reference manifest-following image, and storing the target manifest-following image into a database. Based on the above disclosure, after the target along with the bill image is obtained, firstly, carrying out quality evaluation on the target along with the bill image by using an image quality evaluation model to obtain an image quality evaluation result, then, judging whether the image quality of the target along with the bill image accords with a preset condition or not based on the image quality evaluation result, wherein if not, outputting image shooting correction guide prompt information to prompt a user to carry out correction shooting so as to acquire the target along with the bill image again, and then, repeating the quality evaluation process until the image quality accords with the preset condition, carrying out a subsequent differential transmission flow, namely, obtaining a reference along with th