Automatic verification method and system for image recognition capability

Abstract

The invention provides an automatic verification system for image recognition capability. The system comprises a data management module, a configuration management module, a test execution module anda report display module; various types of image resources are stored as a training set and a test set through a data management module, and a test type corresponding to test execution and required interface information are set through a configuration management module; and the test execution module automatically executes training and testing according to the configuration information, records testinformation in an execution process, and finally generates a corresponding test report through the report display module. The invention further provides an automatic verification method for the imagerecognition capability, the automatic test can be completed, the input time of the test can be greatly reduced, and the manual operation is liberated.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to an automatic verification method and system for image recognition capabilities. Background technique [0002] Today, with the rapid development of the field of intelligence, image recognition is a very important direction, and it is particularly important for the recognition system to have a high enough recognition rate. So how to measure the image recognition ability has become a very important link. At present, face recognition, flower recognition, animal recognition and other scenes are common in the field of image recognition, and the most common test method at present is to verify the recognition effect or ability of the system by manually taking pictures and uploading. The regression verification of automated traversal is performed on the data to verify whether it returns as expected. [0003] The existing image recognition verification technology has the follow...

AI Technical Summary

Problems solved by technology

[0004] Disadvantage 1. If you pass the functional test and acceptance, although it is intuitive, it is very weak when the data volume is high.

It is particularly prominent in the regression test of version iteration or BUG repair. Program changes may only require a few lines of code, but the scope of the test set for verification ranges from hundreds to tens of thousands. A manual test is passed , time-consuming and easy to consume the patience of testers

[0005] Disadvantage 2. At present, the accuracy verification of image recognition is more based on the terminal (client) method, which cannot effectively measure the response of the server in terms of response time.

[0006] Disadvantage 3. Conventional script execution can only verify a recognition effect in the current system environment, including but not limited to recognition results such as similarity rate and response time, but it cannot measure a dynamic performance of the entire system

[0007] Disadvantage 4, conventional image recognition mainly lies in verifying whether the image recognition result is correct, that is, whether the object in the image is recognized

The recognition of an object is based on the similarity rate between the picture and the most similar picture in the current training set. If the threshold is set too high, the corresponding person may not be recognized. If the threshold is set too low, it may cause errors in recognition and misidentification

However, conventional testing methods cannot give suitable adjustment reference suggestions, because there is no data support

Images