CN111597857A - Logistics package detection method, device and equipment and readable storage medium - Google Patents

Abstract

The invention discloses a method for detecting a logistics package, comprising: collecting an image of a package located on a sorting conveyor; a first data matrix consisting of X coordinate values of each coordinate in the image and second data consisting of Y coordinate values The matrix, and the gray value of the image are fused into the image information of the image; the image information is processed by a preset detection model to obtain the category of the target contained in the image information, and the position information of the bounding box covering each target; the image Targets included in the information include at least packages and organ boards. The method can determine the position information of the bounding box covering each target in the image, that is, the position information of each target, and the coordinate information of the image is added in the detection process, thereby improving the position of the logistics package on the sorting conveyor belt. Detection accuracy and package sorting accuracy. The logistics package detection device, equipment and readable storage medium disclosed by the present invention also have the above technical effects.

Description

A kind of logistics package detection method, device, equipment and readable storage medium

technical field

The present invention relates to the technical field of image recognition, and more particularly, to a logistics package detection method, device, equipment and readable storage medium.

Background technique

In recent years, with the rapid development of e-commerce and artificial intelligence technology, the logistics industry has ushered in explosive growth. As a result, many automated and intelligent systems used in logistics scenarios have been born. The logistics package automatic sorting system is based on the basic process of sorting operations, and integrates the core technologies of the Internet of Things such as image recognition, automatic control, and data communication, and realizes the standardized, standardized and intelligent sorting of express packages. It effectively improves the efficiency and quality of express sorting, and promotes the intelligent upgrade of the logistics supply chain.

Figure 1 shows a logistics package sorting system, including: a package supply table, a package detection camera, a barcode identification camera, a plurality of sorting ports and a sorting conveyor belt; the sorting conveyor belt is composed of a trolley and an organ board connected, and every two The trolleys are connected by an organ board. The workflow of the logistics parcel sorting system is as follows: the staff arranges the parcels, puts them face-up, and sends them to the packaging supply table. When there is an empty trolley, the packages on the packaging supply table are sent to the sorting table. Conveyor belt, face sheet is a logistics sheet containing logistics information and identification barcode. When the package is transported by the sorting conveyor to the position of the package detection camera, the package detection camera detects the position of the package on the sorting conveyor, so as to determine whether the position of the package is convenient for the barcode recognition camera to scan; When the package is transported by the sorting conveyor to the barcode When identifying the location of the camera, the barcode recognition camera scans the barcode of the package to determine which sorting port the current package is being delivered to.

It can be seen that the main function of the package detection camera is to determine the position of the package on the sorting conveyor, so as to determine whether the position of the package needs to be adjusted. In the existing package detection cameras, the traditional object detection algorithm is generally used to determine the position of the package. However, because the traditional object detection algorithm relies on artificially designed image features (grayscale, color, texture, etc.) The ability to express features is lacking; at the same time, the traditional target detection algorithm only uses the gray value of the image as the image information of the image, and lacks the ability to express the position information of the image, resulting in the detection accuracy and accuracy of the traditional target detection algorithm. reduced. Therefore, if the traditional object detection algorithm is applied to the package detection camera, the accuracy of the location information of the package detected by the package detection camera will be insufficient, which may lead to the package being delivered to the wrong sorting port.

Therefore, how to improve the position detection accuracy of the logistics package on the sorting conveyor is a problem that needs to be solved by those skilled in the art.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a logistics package detection method, device, equipment and readable storage medium, so as to improve the position detection accuracy of the logistics package on the sorting conveyor belt.

To achieve the above purpose, the embodiments of the present invention provide the following technical solutions:

A logistics package detection method, comprising:

Capture images of packages located on sorting conveyors;

The first data matrix formed by the X coordinate value of each coordinate in the image and the second data matrix formed by the Y coordinate value, and the grayscale value of the image are merged into the image information of the image;

The image information is processed through a preset detection model to obtain the category of the object contained in the image information, and the position information of the bounding box covering each object; wherein, the objects contained in the image information at least include : Wrap and Organ Board.

Wherein, the training steps of the detection model include:

Acquire a training image, and extract the features of two scales of the training image through a convolutional layer and a residual network layer;

The extracted features are fused and analyzed by the feature processing layer to obtain the analysis results;

Determine the detection result of the training image according to the analysis result, and the detection result is: the detection category of the target contained in the training image, and the detection position information of the bounding box covering each target;

Determine whether the difference between the detection result and the labeling information of the training image conforms to a preset condition by using a preset loss function, and the labeling information is the true category of each target included in the training image and the coverage of each target. The true location information of the bounding box;

If so, the detection model training is completed;

Wherein, the residual network layer is composed of multiple network sub-layers of different scales arranged in descending order of scale, and the last two network sub-layers in the residual network layer output two images of the training image respectively. scale features.

Wherein, the features extracted by the feature processing layer are fused and analyzed to obtain analysis results, including:

Convolving and upsampling the first feature output by the penultimate network sublayer in the residual network layer to obtain the second feature output by the penultimate network sublayer in the residual network layer target features of the same scale;

Connect and convolve the target feature and the second feature to obtain a first vector;

Convolving the first feature, and using a vector to represent the convolution result to obtain a second vector;

Parse the first vector and the second vector to obtain the parsing result.

Among them, it also includes:

The loss function is preset according to the difference between the detection result and the annotation information, and the difference between the detection result and the annotation information at least includes: the position error of the bounding box covering the same target, and the bounding box covering the same target. Confidence error with or without target, and class error for the same target.

Wherein, after the acquisition of the training image, the method further includes:

An xml file is used to record the annotation information of the training image.

Wherein, the first data matrix formed by the X coordinate value of each coordinate in the image and the second data matrix formed by the Y coordinate value, and the gray value of the image are fused into the image information of the image Before, also included:

The image is preprocessed, and the preprocessed image is converted into a grayscale image.

Wherein, the described image information is processed by the preset detection model, and the category of the target contained in the image information is obtained, and after covering the position information of the bounding box of each target, it also includes:

Whether the position of the package needs to be adjusted is determined according to the position information of the bounding box covering the package in the image.

A logistics parcel detection device, comprising:

an acquisition module for acquiring images of packages located on the parcel sorting conveyor;

A fusion module, configured to fuse the first data matrix formed by the X coordinate value of each coordinate in the image, the second data matrix formed by the Y coordinate value, and the gray value of the image into an image of the image information;

A detection module, configured to process the image information through a preset detection model, to obtain the category of the target contained in the image information, and the position information of the bounding box covering each target; wherein, in the image information Included targets include at least: wrap and organ board.

A logistics parcel detection equipment, comprising:

memory for storing computer programs;

The processor is configured to implement the steps of the logistics package detection method described in any one of the above when executing the computer program.

A readable storage medium, a computer program is stored on the readable storage medium, and when the computer program is executed by a processor, the steps of the logistics package detection method described in any one of the above are realized.

It can be seen from the above solutions that a method for detecting a logistics package provided by an embodiment of the present invention includes: collecting an image of a package located on a sorting conveyor belt; a first data matrix composed of X coordinate values of each coordinate in the image The second data matrix formed with the Y coordinate value, and the gray value of the image are fused into the image information of the image; the image information is processed by a preset detection model, and the information contained in the image information is obtained. The category of the object, and the position information of the bounding box covering each object; wherein, the objects contained in the image information at least include: a package and an organ board.

It can be seen that, for the collected image of the package located on the sorting conveyor, the method firstly combines the first data matrix formed by the X coordinate value of each coordinate in the image and the second data matrix formed by the Y coordinate value, and the image The gray value of the image is fused into the image information of the image, that is, the position information of the image is embedded in the image information, so as to determine the position information of the target in the image. Further, the image information containing the coordinate information is processed by the preset detection model to obtain the category of the target contained in the image information, and the position information of the bounding box covering each target; wherein, the target contained in the image information is at least Includes: Wrap and Organ Board. In this way, the position information of the bounding box of each target in the image can be determined, that is, the position information of each target, that is, the position information of the logistics package on the sorting conveyor belt can be detected. Therefore, the detection accuracy of the position of the logistics package on the sorting conveyor belt is improved, and the accuracy of the package sorting can be improved.

Correspondingly, a logistics package detection device, equipment, and readable storage medium provided by the embodiments of the present invention also have the above-mentioned technical effects.

Description of drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative efforts.

1 is a schematic diagram of a logistics parcel sorting system disclosed in an embodiment of the present invention;

2 is a flowchart of a method for detecting a logistics package disclosed in an embodiment of the present invention;

3 is a flowchart of a detection model training method disclosed in an embodiment of the present invention;

4 is a schematic diagram of a logistics package detection device disclosed in an embodiment of the present invention;

5 is a schematic diagram of a logistics package detection device disclosed in an embodiment of the present invention;

6 is a schematic diagram of a logistics package detection model disclosed in an embodiment of the present invention;

7 is a schematic diagram of a logistics package analysis result disclosed in an embodiment of the present invention;

8 is a schematic diagram of a logistics package image information disclosed in an embodiment of the present invention;

9 is a schematic diagram of the position information error of a bounding box covering a target in an image disclosed in an embodiment of the present invention;

FIG. 10 is an effect diagram of a production test disclosed in an embodiment of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work, all belong to the protection scope of the present invention.

The embodiments of the present invention disclose a method, device, equipment and readable storage medium for detecting a logistics package, so as to improve the position detection accuracy of the logistics package on the sorting conveyor belt.

Referring to FIG. 2, a method for detecting a logistics package provided by an embodiment of the present invention includes:

S201, collecting images of packages located on the sorting conveyor;

S202, the first data matrix that the X coordinate value of each coordinate in the image is formed of and the second data matrix that the Y coordinate value is formed of, and the grayscale value of the image is merged into the image information of the image;

Wherein, before the first data matrix formed by the X coordinate value of each coordinate in the image, the second data matrix formed by the Y coordinate value, and the gray value of the image are fused into the image information of the image, the method further includes: performing the image processing on the image. Preprocess and convert the preprocessed image to a grayscale image.

S203, processing the image information through a preset detection model to obtain the category of the target contained in the image information, and the position information of the bounding box covering each target.

Among them, the objects included in the image information at least include: a package and an organ board. The location information of the bounding box includes: the length, width and coordinates of the center point of the bounding box, or the coordinates of the four vertices of the bounding box.

Based on the application scenario of the present invention, the present invention only needs to detect whether the location of the package is convenient for scanning by the barcode camera. When the package is in the center of the trolley, this position is ideal, and the scanning angle is the best; when the package is positioned off the center of the trolley, but still on the trolley, the package sorting system can control the trolley to roll so that the package is in the center of the trolley The center of the trolley; and when the package is on the edge of the accordion board or the sorting conveyor, the position of the package cannot be adjusted, which can cause the package to fall into the wrong sorting port or need to manually adjust the position of the package.

Therefore, in this embodiment, after determining the target category contained in the image and the position information of the bounding box covering each target, the number and position of each package and the number and position of the organ boards contained in the current image can be determined. According to the obtained position information, it can be determined whether the position of the package needs to be adjusted and how to adjust the position of the package.

The image information is processed by a preset detection model to obtain the category of the target contained in the image information and the position information of the bounding box covering each target, and further includes: according to the position of the bounding box covering the package in the image information to determine if the package's position needs to be adjusted.

Specifically, determining whether the position of the package needs to be adjusted according to the position information of the bounding box covering the package in the image includes: calculating the area of the bounding box covering each target in the image; and judging whether a part of the package is located outside the sorting conveyor belt.

When a part of the package is not located outside the sorting conveyor, for a package, determine whether there is an overlapping area between the bounding box covering the current package and the bounding box covering the organ board;

If so, calculate the overlapping area M of the bounding box covering the current package and the bounding box covering the organ board, and the sum N of the area covering the bounding box of the current package and the area of the bounding box covering the organ board; and calculate M and N-M When the ratio is greater than the preset threshold, it is considered that the current position of the package needs to be adjusted. At this time, the corresponding adjustment information can be generated and sent to the management terminal for display, so that the staff can manually intervene. Among them, N-M represents the difference between N and M.

Similarly, when a part of the package is located outside the sorting conveyor, for a package, determine whether there is an overlapping area between the bounding box covering the current package and the bounding box covering the outside of the sorting conveyor;

If so, calculate the overlapping area M of the bounding box covering the current package and the bounding box covering the outside of the sorting conveyor belt, and the sum N of the area of the bounding box covering the current package and the area of the bounding box covering the outside of the sorting conveyor belt; and calculate The ratio of M and N-M, when the ratio is greater than the preset threshold, it is considered that the position of the current package needs to be adjusted. At this time, the trolley of the sorting conveyor is controlled to roll so that the current package is located on the trolley. Among them, the outside of the sorting conveyor belt is the position on both sides of the sorting conveyor belt.

It can be seen that this embodiment provides a logistics package detection method. For the collected image of the package located on the sorting conveyor belt, the method firstly combines the first data matrix composed of the X coordinate value of each coordinate in the image and the The second data matrix formed by the Y coordinate value and the gray value of the image are fused into the image information of the image, that is, the position information of the image is embedded in the image information, so as to determine the position information of the target in the image. Further, the image information containing the coordinate information is processed by the preset detection model to obtain the category of the target contained in the image information, and the position information of the bounding box covering each target; wherein, the target contained in the image information is at least Includes: Wrap and Organ Board. In this way, the position information of the bounding box of each target in the image can be determined, that is, the position information of each target, that is, the position information of the logistics package on the sorting conveyor belt can be detected. Therefore, the detection accuracy of the position of the logistics package on the sorting conveyor belt is improved, and the accuracy of the package sorting can be improved.

Referring to Figure 3, the training steps of the detection model include:

S301. Obtain a training image, and extract features of two scales of the training image through a convolution layer and a residual network layer;

S302, fuse the extracted features through the feature processing layer and analyze them to obtain an analysis result;

S303, determining the detection result of the training image according to the analysis result;

Among them, the detection result is: the detection category of the target contained in the training image, and the detection position information of the bounding box covering each target;

S304, judge whether the difference between the detection result and the labeling information of the training image by the preset loss function meets the preset condition; if so, execute S305; if not, execute S306;

Among them, the annotation information is the real category of each target contained in the training image and the real position information of the bounding box covering each target;

S305, the detection model training is completed;

S306 , update the parameters of the detection model through the stochastic gradient descent method, and execute S301 .

Among them, the residual network layer is composed of multiple network sub-layers of different scales arranged from large to small, and the last two network sub-layers in the residual network layer respectively output the features of the two scales of the training image.

Among them, the extracted features are fused and analyzed by the feature processing layer, and the analysis results are obtained, including:

Convolve and upsample the first feature output by the penultimate network sublayer in the residual network layer to obtain a target with the same scale as the second feature output by the penultimate network sublayer in the residual network layer feature;

Connect and convolve the target feature and the second feature to obtain the first vector;

Convolve the first feature, and use a vector to represent the convolution result to obtain a second vector;

Parse the first vector and the second vector to get the parsing result.

Among them, it also includes:

The loss function is preset according to the difference between the detection result and the annotation information. The difference between the detection result and the annotation information includes at least: the position error of the bounding box covering the same target, the confidence error of whether there is a target in the bounding box covering the same target, and the same The class error of the target.

Wherein, after acquiring the training image, the method further includes: recording the annotation information of the training image by using an xml file. That is: record the true category of each object in the image and the true location information of the bounding box covering each object.

The following describes a logistics package detection device provided by the embodiments of the present invention. The logistics package detection device described below and the logistics package detection method described above can be referred to each other.

Referring to FIG. 4 , a logistics package detection device provided by an embodiment of the present invention includes:

The collection module 401 is used to collect the image of the package located on the package sorting conveyor;

The fusion module 402 is used to fuse the first data matrix formed by the X coordinate value of each coordinate in the image, the second data matrix formed by the Y coordinate value, and the gray value of the image into the image of the image. image information;

The detection module 403 is configured to process the image information through a preset detection model to obtain the category of the target contained in the image information, and the position information of the bounding box covering each target; wherein, the image information Included targets include at least: Wraps and Organ Boards.

Among them, it also includes: a training module for training the detection model, including:

an acquisition unit for acquiring a training image, and extracting features of two scales of the training image through a convolutional layer and a residual network layer;

The parsing unit is used to fuse and parse the extracted features through the feature processing layer to obtain parsing results;

A determination unit, for determining the detection result of the training image according to the analysis result, the detection result is: the detection category of the target included in the training image, and the detection position information of the bounding box covering each target;

A judging unit, configured to judge whether the difference between the detection result and the labeling information of the training image conforms to a preset condition through a preset loss function, where the labeling information is the true category of each target included in the training image and the ground-truth location information of the bounding box covering each object;

Completion unit, when the difference between the labeling information of the detection result and the training image meets a preset condition, the training of the detection model is completed;

Wherein, the residual network layer is composed of multiple network sub-layers of different scales arranged in descending order of scale, and the last two network sub-layers in the residual network layer output two images of the training image respectively. scale features.

Wherein, the parsing unit includes:

The sampling subunit is used to convolve and upsample the first feature output by the penultimate network sublayer in the residual network layer to obtain the same value as the penultimate network sublayer in the residual network layer. The second feature output of the layer has the same target feature scale;

a first convolution subunit, configured to connect and convolve the target feature and the second feature to obtain a first vector;

The second convolution subunit is used to convolve the first feature, and represents the convolution result with a vector to obtain a second vector;

A parsing subunit, configured to parse the first vector and the second vector to obtain the parsing result.

Wherein, the training module also includes:

A preset unit, configured to preset the loss function according to the difference between the detection result and the annotation information, where the difference between the detection result and the annotation information at least includes: covering the position error of the bounding box of the same target, covering The confidence error of whether there is a target in the bounding box of the same target, and the class error of the same target.

Wherein, the training module also includes:

The recording unit is used for recording the annotation information of the training image by using the xml file.

Among them, it also includes:

The preprocessing unit is used to preprocess the image and convert the preprocessed image into a grayscale image.

Among them, it also includes:

A determining module, configured to determine whether the position of the package needs to be adjusted according to the position information of the bounding box covering the package in the image.

It can be seen that this embodiment provides a logistics package detection device, including: a collection module, a fusion module and a detection module. First, the acquisition module collects the image of the package located on the package sorting conveyor; then the fusion module combines the first data matrix formed by the X coordinate value of each coordinate in the image and the second data matrix formed by the Y coordinate value, and The gray value of the image is fused into the image information of the image; finally, the detection module processes the image information through a preset detection model, obtains the category of the target contained in the image information, and covers each target. The location information of the bounding box; wherein, the objects contained in the image information at least include: a package and an organ board. In this way, the various modules work in division of labor and perform their own duties, thereby improving the detection accuracy of the location of the logistics package on the sorting conveyor belt and the accuracy of package sorting.

A kind of logistics parcel detection equipment provided by the embodiment of the present invention is introduced below, a kind of logistics parcel detection equipment described below and a kind of logistics parcel detection method and device described above can be referred to each other.

Referring to FIG. 5 , a logistics package detection device provided by an embodiment of the present invention includes:

a memory 501 for storing computer programs;

The processor 502 is configured to implement the steps of the logistics package detection method described in any of the above embodiments when executing the computer program.

A readable storage medium provided by an embodiment of the present invention will be introduced below. A readable storage medium described below and a logistics package detection method, device, and device described above can be referred to each other.

A readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, implements the steps of the method for detecting a logistics package as described in any of the foregoing embodiments.

Based on the detection model training steps provided by the present invention, the following detection models can be designed, please refer to FIG. 6 . The detection model shown in Figure 6 includes: a feature extraction module, a feature fusion module and a parsing module, wherein:

The feature extraction module includes a convolution layer and five residual network blocks, and the feature scales extracted by each residual network block are different;

The feature fusion module is used to convolve the features output by the last residual network block, and then upsample it to obtain a feature with a scale of "26*26*256", which is combined with the penultimate residual network. The features of the block output are connected to obtain features with a scale of "26*26*768", and the features with a scale of "26*26*768" are convolved to obtain a feature with a scale of "13*13*1024";

The feature parsing module is used to convolve two features with a scale of "13*13*1024" respectively, and parse the two convolution results obtained, thereby outputting a data block.

Among them, the two convolution results obtained are analyzed, that is, to determine the number and category of objects contained in the current image, and to cover the position information of the bounding box of each object, etc.

Among them, the feature size of the final output of the feature extraction module is "13*13*1024", and the size of the output of the previous layer is "26*26*512", and these two features are input into the feature fusion module. Among them, the features of "13*13*1024" are divided into two paths. One path goes through 3x3 convolution and 1x1 convolution to get "13*13*21"; the other path goes through the upsampling layer to expand the data block into "26*26*" 256", then fuse with the penultimate feature 26*26*512 to get "26*26*768", and connect the upsampling result with "26*26*512" to get "26*26*768", Convolve the feature of "26*26*768" to get the feature with the scale of "13*13*1024", and the feature of "13*13*1024" will undergo 3x3 convolution and 1x1 convolution to get "26*26" *twenty one". In this way, we finally get two results: "26*26*21" and "13*13*21".

It should be noted that since the size of the small package and the large package is quite different, and the size of the organ board is much larger than that of the general package, the feature extraction module has two output scales, 13x13 and 26x26 respectively. Among them, the 13x13 scale is used to detect large packages and organ boards, and the 26x26 scale is used to detect small packages.

The two features "26*26*21" and "13*13*21" go through the feature analysis module, and the feature analysis module will extract the category information, coordinate information and the probability of the existence of the target in the bounding box from the data block. For example: three bounding boxes are predicted, namely box1, box2 and box3, and the information contained in each bounding box is: x and y represent the abscissa and ordinate of the center position of the bounding box, respectively, w and h represent the boundary The length and width of the box, p represents the probability of the existence of the target in the bounding box, and class1 and class2 represent the category of the target at this location. The analysis results are shown in Figure 7.

It should be noted that, before the image is input into the feature extraction module, the first data matrix composed of the X coordinate value of each coordinate in the image, the second data matrix composed of the Y coordinate value, and the gray value of the image are fused into the image. image information to embed location information in the image information. Please refer to Figure 8 for specific image information.

In order to measure the error between the detection result contained in the output data block and the real result, that is: the error between the position information of the bounding box covering target A in the detection result and the position information of the bounding box covering target A in the real result; the detection result The error between the confidence of whether there is a target in the bounding box covering target A, and the confidence of whether there is a target in the bounding box covering target A in the real result; the category of target A determined in the detection result is different from the target in the real result. A's class of error. The present invention presets the loss function according to the differences in the above three aspects, specifically:

1. The error of the position information of the bounding box is represented by formula (1):

表示 第i个网格中的第j个边界框是否覆盖当前目标，其中，与当前目标的真实边 界框的Iou最大的边界框为覆盖当前目标的预测边界框，请参见图9，其中 B＝3，预测位置即为检测结果中覆盖目标A的边界框的位置信息，x’和y’为边 界框的中心点坐标，w’为边界框的长，h’为边界框的宽；真实位置即为标注信 息中覆盖目标A的边界框的位置信息，x和y为边界框的中心点坐标，w为 边界框的长，h为边界框的宽。where S represents the number of grids divided in the image, B represents the number of predicted bounding boxes,

Indicates whether the jth bounding box in the ith grid covers the current target, wherein the bounding box with the largest Iou of the true bounding box of the current target is the predicted bounding box covering the current target, see Figure 9, where B= 3. The predicted position is the position information of the bounding box covering the target A in the detection result, x' and y' are the coordinates of the center point of the bounding box, w' is the length of the bounding box, and h' is the width of the bounding box; the real position That is, the position information of the bounding box covering the target A in the annotation information, x and y are the coordinates of the center point of the bounding box, w is the length of the bounding box, and h is the width of the bounding box.

2. The error of confidence is expressed by formula (2):

in,

3. The error of the category is represented by formula (3), and the categories are two types: package and organ board:

Combining 1, 2 and 3, the resulting loss function is formula (4):

Loss=Loss _coord +Loss _conf +Loss _class (4)

In order to prove the superiority of the present invention, different detection methods are used to detect the same batch of data, and the detection results are shown in Table 1.

Table 1

As can be seen from Table 1, Faster R-CNN has higher accuracy but slower processing speed; SSD, Yolo v3 and tiny Yolo have slightly lower accuracy but faster processing speed. The accuracy rate of the present invention is the closest to Faster R-CNN, and the processing speed is the fastest at the same time. It can be seen that the present invention not only ensures the accuracy rate, but also improves the processing efficiency.

The invention has also been tested in an actual production environment. 28,209 sheets were randomly selected from all the test result graphs collected within one month. Through manual inspection, only 3 result graphs missed a package. That is to say, the false detection rate of the present invention is only one in ten thousand. Some renderings of the production test are shown in Figure 10, and these images are from three different sorting lines.

The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the various embodiments can be referred to each other.

The above description of the disclosed embodiments enables any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (10)

1. a logistics parcel detection method, is characterized in that, comprises: Capture images of packages located on sorting conveyors; The first data matrix formed by the X coordinate value of each coordinate in the image and the second data matrix formed by the Y coordinate value, and the gray value of the image are merged into the image information of the image; The image information is processed through a preset detection model to obtain the category of the object contained in the image information, and the position information of the bounding box covering each object; wherein, the objects contained in the image information at least include : Wrap and Organ Board. 2. The logistics package detection method according to claim 1, wherein the training step of the detection model comprises: Acquire a training image, and extract features of two scales of the training image through a convolutional layer and a residual network layer; The extracted features are fused and analyzed by the feature processing layer to obtain the analysis results; The detection result of the training image is determined according to the analysis result, and the detection result is: the detection category of the target contained in the training image, and the detection position information of the bounding box covering each target; Determine whether the difference between the detection result and the labeling information of the training image conforms to a preset condition by using a preset loss function, and the labeling information is the true category of each target included in the training image and the coverage of each target. The true location information of the bounding box; If so, the detection model training is completed; Wherein, the residual network layer is composed of multiple network sub-layers of different scales arranged in descending order of scale, and the last two network sub-layers in the residual network layer output two images of the training image respectively. scale features. 3. The logistics package detection method according to claim 2, wherein the features extracted by the feature processing layer are fused and analyzed to obtain analysis results, comprising: Convolving and upsampling the first feature output by the penultimate network sublayer in the residual network layer to obtain the second feature output by the penultimate network sublayer in the residual network layer target features of the same scale; Connect and convolve the target feature and the second feature to obtain a first vector; Convolving the first feature, and using a vector to represent the convolution result to obtain a second vector; Parse the first vector and the second vector to obtain the parsing result. 4. logistics parcel detection method according to claim 3, is characterized in that, also comprises: The loss function is preset according to the difference between the detection result and the annotation information, and the difference between the detection result and the annotation information at least includes: the position error of the bounding box covering the same target, and the bounding box covering the same target. Confidence error with or without target, and class error for the same target. 5. The logistics package detection method according to claim 4, characterized in that, after the acquisition of the training image, further comprising: An xml file is used to record the annotation information of the training image. 6. The logistics package detection method according to any one of claims 1-5, wherein the first data matrix formed by the X coordinate value of each coordinate in the image and the Y coordinate value are formed by The second data matrix, and before the gray value of the image is fused into the image information of the image, further includes: The image is preprocessed, and the preprocessed image is converted into a grayscale image. 7. The logistics package detection method according to any one of claims 1-5, wherein the image information is processed through a preset detection model to obtain the category of the target contained in the image information , and after the location information of the bounding box covering each target, it also includes: Whether the position of the package needs to be adjusted is determined according to the position information of the bounding box covering the package in the image. 8. A logistics parcel detection device, characterized in that, comprising: an acquisition module for acquiring images of packages located on the parcel sorting conveyor; A fusion module, configured to fuse the first data matrix formed by the X coordinate value of each coordinate in the image, the second data matrix formed by the Y coordinate value, and the gray value of the image into an image of the image information; a detection module, configured to process the image information through a preset detection model to obtain the category of the target contained in the image information, and the position information of the bounding box covering each target; wherein, the image information contains Included targets include at least: wrap and organ board. 9. A logistics parcel detection device, characterized in that, comprising: memory for storing computer programs; The processor is configured to implement the steps of the logistics package detection method according to any one of claims 1-7 when executing the computer program. 10. A readable storage medium, characterized in that, a computer program is stored on the readable storage medium, and when the computer program is executed by a processor, the logistics package detection according to any one of claims 1-7 is realized steps of the method.

Images