reference
The basic model: https://blog.csdn.net/program_developer/article/details/80369989
If a class implements: https://www.cnblogs.com/Mrzhang3389/p/9790955.html
Seeking precision, tensorboard Visualization: https://blog.csdn.net/wwxy1995/article/details/80832283
https://blog.csdn.net/lansetiankong2104/article/details/79981776
These two like, comment, and loss of function is slightly different
Understanding of the data set: https://blog.csdn.net/qq_41185868/article/details/79094752
minst dataset Visualization: https://blog.csdn.net/wc13197389627/article/details/95527289
Save and load model: https://blog.csdn.net/thriving_fcl/article/details/71423039
How to identify your own handwriting (ps) and mnist digital data set is a Considerations such as the West handwriting: https://blog.csdn.net/qq_38269418/article/details/78991649
Model building, complete tutorial training, recognition of handwritten numbers: https://blog.csdn.net/qq_33414271/article/details/86488960
I used one of the pictures preprocessing functions, provided that the first shot, so digital photos fully revealed, and then adjust the threshold value of the two
**** improve the accuracy of parameter adjustment with respect to such methods : https://blog.csdn.net/hu_hao/article/details/95535859
Loading and saving models, according to the model prediction: https://blog.csdn.net/tsfx051435adsl/article/details/88711493
In short, study notes written their own merits.