Lesson 11 Deep Learning on Image Recognition
There were studies to let PCs detect what are captured in pictures. Those are on a field of artificial intelligence, and many results has been published for a long time. Suppose we have many pictures with animals. Some of them may have difficulty in detection even for human beings. It was a difficult job for conventional technologies to recognize those pictures with high accuracy. We now have some results Deep Learning allows us to increase the accuracy of image recognition considerably. The following is a brief explanation of image recognition process with Deep Learning.
1 Learning Data Set Preparation
We collect image files for the system’s learning. For example, a lot of images with dogs or cats should be collected for dog-cat recognition system. And also, we prepare a data with expected results for every image. Collection of data on expected answers is called Teaching Data. Collection of image files and teaching data is called Learning Data Set.
2 Learning Execution
The system is modifying its own parameters, so that the difference between the output and the teaching data is getting less with given learning data set. Large number of image files and teaching data for repetitive calculation results in closer output from any image file.
3 Result Evaluation
We get the result from image files for testing. The data set for this purpose is called Test Data Set. You can confirm the image recognition accuracy after learning.
