Search

CN-115690563-B - Security check machine detection model and training data generation method and device thereof

CN115690563BCN 115690563 BCN115690563 BCN 115690563BCN-115690563-B

Abstract

The application provides a security inspection machine detection model and a training data generation method and device thereof. The method comprises the steps of inputting high-low energy data of a first background object and high-low energy data of a first injected object into a feature extraction model corresponding to a target security inspection machine to obtain high-order substance features of the first background object and high-order substance features of the first injected object, wherein the high-order substance features are used for reflecting more than two substance properties of the object, fusing the high-order substance features of the first background object and the first injected object to obtain fusion features, inputting the fusion features into a feature reduction model corresponding to the target security inspection machine to obtain high-low energy data of the injected object, fusing the first injected object to the first background object, obtaining a first original image according to the high-low energy data of the injected object, and coloring the first original image to obtain a first training image. The method can generate training data for improving the detection effect of the small target with low cost and high efficiency.

Inventors

  • WANG FENG

Assignees

  • 熵基科技股份有限公司

Dates

Publication Date
20260505
Application Date
20221117

Claims (10)

  1. 1. The training data generation method of the security inspection machine detection model is characterized by comprising the following steps of: acquiring high-low energy data of a first background object and a first injected object, wherein the high-low energy data comprises high-energy data and low-energy data; Respectively inputting the high-low energy data of the first background object and the high-low energy data of the first injected object into a feature extraction model corresponding to a target security inspection machine to obtain a high-order substance feature of the first background object and the high-order substance feature of the first injected object, wherein the high-order substance feature is used for reflecting more than two substance attributes of the object; Fusing the high-order substance characteristics of the first background object and the first injection object to obtain fusion characteristics; inputting the fusion characteristics into a characteristic reduction model corresponding to the target security inspection machine to obtain the high-low energy data of the injected article, wherein the injected article is the result of fusing the first injected article to the first background article; obtaining a first original image according to the high-low energy data of the injected article; and coloring the first original image to obtain a first training image.
  2. 2. The method of claim 1, wherein the training data generating method further comprises, after obtaining a plurality of the first training images: Acquiring high-low energy imaging of a second background object and a second injected object; And inquiring the pixel values corresponding to the pixel values of the group of pixel points according to the pixel values of the group of pixel points and an injection mapping table, and determining the pixel values of the pixel points of a second training image at the same position according to the inquired pixel values to obtain the second training image, wherein the injection mapping table is used for reflecting the mapping relation between the pixel values of the pixel points in the high-low energy imaging of the first background object and the first injection object and the pixel values of the pixel points at the same position in the corresponding first training image.
  3. 3. The method of claim 1, wherein the coloring the first original image comprises: And selecting a coloring mode corresponding to the target security inspection machine to color the first original image.
  4. 4. The method of claim 1, further comprising, after obtaining the first training image: And carrying out data enhancement on the first training images to obtain a plurality of third training images.
  5. 5. The method of claim 4, wherein the manner of data enhancement comprises any one of: geometrically transforming the object in the first training image; and performing color transformation on the objects in the first training image.
  6. 6. The security inspection machine detection model product is characterized by comprising a backbone network, a feature fusion network and a detection network which are sequentially connected, wherein the security inspection machine detection model is obtained by training a training set obtained by the training data generation method according to any one of claims 1-5.
  7. 7. The security inspection machine detection model product of claim 6, wherein the tail of the feature fusion network is configured with ASFF modules.
  8. 8. The utility model provides a training data generation device of security inspection machine detection model which characterized in that includes: The high-low energy data acquisition module is used for acquiring high-low energy data of the first background object and the first injected object, wherein the high-low energy data comprises high-energy data and low-energy data; The characteristic extraction module is used for respectively inputting the high-low energy data of the first background article and the high-low energy data of the first injection article into a characteristic extraction model corresponding to a target security inspection machine to obtain a high-order substance characteristic of the first background article and the high-order substance characteristic of the first injection article, wherein the high-order substance characteristic is used for reflecting more than two substance attributes of the articles; The feature fusion module is used for fusing the high-order substance features of the first background object and the first injection object to obtain fusion features; The feature reduction module is used for inputting the fusion features into a feature reduction model corresponding to the target security inspection machine to obtain the high-low energy data of the injected article, wherein the injected article is a result of fusing the first injected article to the first background article; The imaging module is used for obtaining a first original image according to the high-low energy data of the injected article; And the coloring module is used for coloring the first original image to obtain a first training image.
  9. 9. A storage medium having stored therein computer readable instructions which, when executed by one or more processors, cause the one or more processors to perform the steps of the training data generation method of any of claims 1 to 5.
  10. 10. A computer device comprising one or more processors and a memory having stored therein computer readable instructions which, when executed by the one or more processors, perform the steps of the training data generation method of any of claims 1 to 5.

Description

Security check machine detection model and training data generation method and device thereof Technical Field The application relates to the technical field of security inspection machines, in particular to a security inspection machine detection model, a training data generation method and device thereof, a storage medium and computer equipment. Background Safety inspection is an important means for finding and eliminating accident potential, implementing safety measures and preventing accidents. In recent years, X-ray machines have been widely used for security inspection in public places such as subway stations and airports. The security check personnel mainly check the image that X ray security check camera was shot with naked eyes, leads to easily leaking to examine and the false detection. In order to improve efficiency, the method also appears that a security inspection machine detection model can be established based on deep learning, and automatic detection is realized through the security inspection machine detection model. However, the detection model of the security inspection machine is driven by data, the quantity and the quality of training sets determine the detection precision, and the traditional training data generation method cannot ensure the small target detection effect of the detection model of the security inspection machine. Disclosure of Invention The application aims to at least solve one of the technical defects, and particularly the technical defects that training images in the prior art are difficult to help a security inspection machine detection model to improve the detection effect of a small target. In a first aspect, an embodiment of the present application provides a training data generating method for a security inspection machine detection model, including: Acquiring high-low energy data of a first background object and a first injected object; Respectively inputting the high-low energy data of the first background object and the high-low energy data of the first injection object into a feature extraction model corresponding to the target security inspection machine to obtain a high-order substance feature of the first background object and a high-order substance feature of the first injection object, wherein the high-order substance feature is used for reflecting more than two substance attributes of the object; fusing the high-order substance characteristics of the first background object and the first injection object to obtain fusion characteristics; inputting the fusion characteristics into a characteristic reduction model corresponding to a target security inspection machine to obtain high-low energy data of the injected articles, wherein the injected articles are the result of fusing the first injected articles to the first background articles; Obtaining a first original image according to the high-low energy data of the injected object; and coloring the first original image to obtain a first training image. In one embodiment, after obtaining the plurality of first training images, the training data generating method further includes: Acquiring high-low energy imaging of a second background object and a second injected object; And inquiring the pixel values corresponding to the pixel values of the group of pixel points according to the pixel values of the group of pixel points and an injection mapping table, and determining the pixel values of the pixel points of the second training image at the same position according to the inquired pixel values to obtain the second training image, wherein the injection mapping table is used for reflecting the mapping relation between the pixel values of each pixel point in the high-low energy imaging of the first background object and the first injection object and the pixel values of the pixel points at the same position in the corresponding first training image. In one embodiment, coloring the first original image includes: and selecting a coloring mode corresponding to the target security inspection machine to color the first original image. In one embodiment, after obtaining the first training image, the method further includes: and carrying out data enhancement on the first training images to obtain a plurality of third training images. In one embodiment, the manner of data enhancement includes any one of the following: geometrically transforming the object in the first training image; The items in the first training image are color transformed. In a second aspect, an embodiment of the present application provides a security inspection machine detection model, including a backbone network, a feature fusion network, and a detection network that are sequentially connected, where the security inspection machine detection model is obtained by training a training set obtained by using the training data generating method in any one of the foregoing embodiments. In one embodiment, the tail of the feature fusion network is confi