KR-102962621-B1 - APPRATUS FOR ANALYSING IMAGE AND METHOD THEREOF

Abstract

The present disclosure relates to an image analysis method for anomaly detection and an electronic device for performing the same. The image analysis performed by the electronic device may include: acquiring a first image; generating a second image by auto-encoding the first image; extracting a first feature vector and a second feature vector from each of the first image and the second image; filtering the first feature vector and the second feature vector, respectively, using a filtering vector generated based on element-wise distance values of the first feature vector and the second feature vector; and determining whether the first image is anomaly based on element-wise distance values of the filtered first feature vector and the second feature vector.

Inventors

• 정재일
• 장준익
• 홍종희

Assignees

• 삼성전자주식회사

Dates

Publication Date

20260507

Application Date

20190919

Claims (20)

1. In a method for analyzing images of an electronic device, Step of acquiring a first image; A step of generating a second image by auto-encoding the first image; A step of extracting a first feature vector and a second feature vector from each of the first image and the second image; A step of filtering the first feature vector and the second feature vector, respectively, using a filtering vector generated based on element-wise distance values of the first feature vector and the second feature vector; and A method comprising the step of determining whether the first image is anomaly based on the element-wise distance values of the filtered first feature vector and the second feature vector.
2. delete
3. delete
4. In paragraph 1, the filtering step A step of determining a filtering vector of the same size as the first feature vector and the second feature vector based on element-wise distance values of the first feature vector and the second feature vector; and A method comprising the step of element-wise multiplying the determined filtering vector with the first feature vector and the second feature vector.
5. In paragraph 4, the step of determining the filtering vector A step of generating the filtering vector including the element-wise distance values of the first feature vector and the second feature vector as element values of the filtering vector; A step of adjusting the element values within the filtering vector based on whether the element values within the generated filtering vector exceed a preset first threshold; and A method comprising the step of determining the filtering vector including the adjusted element values as element values of the filtering vector.
6. delete
7. delete
8. delete
9. delete
10. delete
11. In an electronic device for analyzing images, Memory for storing one or more instructions; and Includes at least one processor; and The above at least one processor executes the above one or more instructions, Acquire the first image, A second image is generated by auto-encoding the first image above, and A first feature vector and a second feature vector are extracted from each of the first image and the second image, and Filtering the first feature vector and the second feature vector, respectively, using a filtering vector generated based on the element-wise distance values of the first feature vector and the second feature vector, and An electronic device that determines whether the first image is anomaly based on the element-wise distance values of the filtered first feature vector and second feature vector.
12. delete
13. delete
14. In paragraph 11, the at least one processor executes the one or more instructions, Based on the element-wise distance values of the first feature vector and the second feature vector, the filtering vector of the same size as the first feature vector and the second feature vector is determined, and An electronic device characterized by multiplying the determined filtering vector above by the first feature vector and the second feature vector element-wise.
15. In paragraph 11, the above at least one processor executes the above one or more instructions, A filtering vector is generated that includes the element-wise distance values of the first feature vector and the second feature vector as element values of the filtering vector, and Adjusting the element values within the filtering vector based on whether the element values within the generated filtering vector exceed a preset first threshold value, and An electronic device for determining the filtering vector that includes the above-mentioned adjusted element values as element values of the filtering vector.
16. In paragraph 15, the above at least one processor executes the above one or more instructions, Based on whether the sum of all element values within the filtering vector is less than or equal to a second threshold, the element values of the filtering vector are readjusted, and An electronic device for determining the filtering vector that includes the above-mentioned readjusted element values as element values within the filtering vector.
17. In paragraph 15, the above at least one processor executes the above one or more instructions, Among the element values within the filtering vector generated above, element values exceeding the first threshold value are adjusted to 0 (zero-element), and An electronic device characterized by adjusting element values within the generated filtering vector that are less than or equal to the first threshold value to 1 (non-element).
18. In paragraph 15, the above at least one processor executes the above one or more instructions, Acquiring a plurality of training images and a plurality of reconstructed images corresponding to each of the plurality of training images, and Element-wise distance values of feature vectors extracted from each of the plurality of training images and the plurality of reconstructed images corresponding to the plurality of training images are summed, and An electronic device characterized by generating the filtering vector that includes the summed element-specific distance values as element values.
19. delete
20. Step of acquiring a first image; A step of generating a second image by auto-encoding the first image; A step of extracting a first feature vector and a second feature vector from each of the first image and the second image; A step of filtering the first feature vector and the second feature vector, respectively, using a filtering vector generated based on element-wise distance values of the first feature vector and the second feature vector; and A computer-readable recording medium storing a program that enables the step of determining whether the first image is anomaly based on the element-wise distance values of the filtered first feature vector and the second feature vector.

Description

Appraisal method for image analysis and electronic device for performing the same {APPRATUS FOR ANALYSING IMAGE AND METHOD THEREOF} The present disclosure relates to a method and apparatus for analyzing an image. More specifically, it relates to an image analysis method and apparatus for anomaly detection. Recently, technologies for detecting defects in products produced in automated manufacturing processes, as well as defects in specific processes or equipment that may occur during the manufacturing process, are being actively researched. Anomaly detection, a data classification technique used to detect abnormal data and products exhibiting patterns different from those of normal products, is widely used to detect defects that may occur during the product manufacturing process. Furthermore, in addition to the field of data management regarding product manufacturing processes, the demand for anomaly detection technology is steadily increasing in areas such as system operation and security-related systems. Generally, in high-yield manufacturing processes, there is a problem in that it is difficult to obtain sufficient abnormal data because the defect rate is low. Therefore, unlike deep-learning-based algorithms that require both normal and abnormal data, anomaly detection is mainly used in high-yield manufacturing processes where it is difficult to obtain sufficient abnormal data due to low defect rates, as it can detect abnormal data based on normal data. However, general anomaly detection has limitations in that it is difficult to obtain a discriminative reconstruction error, as it utilizes only normal data and directly uses the general features extracted from both the input data and the reconstructed data output by inputting the input data into an autoencoder. Furthermore, conventional anomaly detection merely improves the performance of the autoencoder itself, but has the problem of failing to utilize the diverse features output from both the input data and the reconstruction data produced by inputting the said input data into the autoencoder in detecting anomalies. Therefore, in anomaly detection using only normal data, there is a need for the development of methods to measure discriminative reconstruction errors and technologies to utilize features output from both the input data and the reconstruction data. FIG. 1 is a diagram illustrating an image analysis method of an electronic device for anomaly detection according to one embodiment. FIG. 2 is a diagram illustrating the process of determining a reconstruction error in an electronic device according to one embodiment. FIG. 3 is a diagram illustrating an image analysis process for detecting abnormalities in an electronic device according to one embodiment. FIG. 4 is a diagram illustrating the process of an electronic device according to one embodiment filtering a feature vector using a filtering vector. FIG. 5 is a flowchart of an image analysis method of an electronic device according to one embodiment. FIG. 6 is a diagram for specifically explaining how an electronic device according to one embodiment determines a filtering vector. FIG. 7 is a diagram for specifically explaining how an electronic device according to one embodiment determines a filtering vector. FIG. 8 is a drawing for specifically explaining a method for an electronic device according to one embodiment to determine whether there is an abnormality in a first image. FIG. 9 is a diagram illustrating a method for an electronic device according to one embodiment to analyze an image for abnormality detection. FIG. 10 is a diagram illustrating an image analysis process for detecting anomalies by having an electronic device and a server interact with each other according to one embodiment. FIG. 11 is a diagram illustrating the process of performing an image analysis method for anomaly detection by linking an electronic device, an edge computing server, and a main cloud server according to one embodiment. FIG. 12 is a diagram illustrating the process of performing an image analysis method for anomaly detection by interconnecting an electronic device, IoT devices, and an edge computing server according to one embodiment. FIG. 13 is a block diagram of an electronic device for analyzing an image according to one embodiment. FIG. 14 is a block diagram of an electronic device for analyzing an image according to one embodiment. FIG. 15 is a block diagram of a server performing a method for analyzing an image according to one embodiment. FIG. 16 is a drawing for illustrating normal product images and defective product images analyzed by an electronic device according to one embodiment. FIG. 17 is a diagram illustrating the average accuracy indicating how accurately an electronic device (1000) can determine an anomaly in an image. The terms used in this specification will be briefly explained, and the present disclosure will be described in detail. The terms used in this disclosure have been selec