You can read the analog meter with the example MNIST.
Part 1 https://qiita.com/bono0/items/bf9ec40d0f63d7a193b1
https://bono0.com/
When reading an analog meter, most of the prediction accuracy is determined in the process before prediction by artificial intelligence. There is a limit to what can be done by adjusting the artificial intelligence itself. First of all, proper shooting is important. Artificial intelligence cannot make a judgment on images that cannot be judged by humans. Next is image preprocessing. Since the direction of the needle of the analog meter is especially important, perform the same preprocessing as the convolutional neural network (CNN), which has a good reputation for image processing. The only difference is whether to fold it manually or let artificial intelligence fold it.
By pretreatment, you can freely emphasize or erase according to the direction of the needle. By pre-processing the vertical, horizontal and diagonal convolutions, the accuracy of artificial intelligence is greatly improved.
![output03[1].png](https://qiita-image-store.s3.ap-northeast-1.amazonaws.com/0/237930/b4c3fae7-533f-2a03-93a5-ad714a661338.png)
![output02[1].png](https://qiita-image-store.s3.ap-northeast-1.amazonaws.com/0/237930/6b247628-7fa3-4f57-0221-c18fd2517ada.png)
![dx05[1].png](https://qiita-image-store.s3.ap-northeast-1.amazonaws.com/0/237930/a1f9e4b4-16bf-f2a1-43fd-8e5befe3adba.png)
Recommended Posts