NOTE: This is an indea, not a solution. You can use any algorithm to extract features or classificate.
CNN always output a result while given an image. So I am thinking of deploying a pre-classification algorithm to decide wether throw the image into CNN or give out the result.
Content of train_all.txt and test_all.txt is: image path + space + category index
特征提取&分类博客中文地址：http://blog.csdn.net/renhanchi/article/details/79127914
寻找最优参数博客中文地址：http://blog.csdn.net/renhanchi/article/details/79139415