实验：是否图片的重叠区域携带了决定分类的所有信息？

最新推荐文章于 2024-05-29 18:03:42 发布

黑榆

最新推荐文章于 2024-05-29 18:03:42 发布

阅读量4.8k

点赞数

分类专栏：用神经网络模拟原子文章标签：深度学习神经网络数据挖掘重叠区域重叠法

本文链接：https://blog.csdn.net/georgesale/article/details/104482525

版权

用神经网络模拟原子专栏收录该内容

358 篇文章 5 订阅

订阅专栏

使用重叠法将参与分类训练的同一批次图片的所有不重叠部分变成0，只保留图片的重叠部分训练网络，如果分类准确率上升表明重叠部分已经包含决定分类的全部信息。如果这个假设成立，表明神经网络实现分类是通过在分类对象之间建立一个公共的重叠区，并通过识别重叠区来实现的。

具体实验过程

mnist
A	B	C
0	1
0	2
0	3
0	4
0	5
0	6
0	7
0	8
0	9
0	1	2
0	1	3

本次实验构造了9个二分类网络和2个三分类网络来验证这个假设是否成立。网络结构是

(A,B)-81*30*2-(1,0)(0,1)

(A,B,C)-81*30*3-(1,00)(0,1,0)(0,0,1)

用一个三层网络来实现分类，不加卷积核，通过同样的固定收敛标准，199次测量取平均值的办法来比较。

收敛标准从0.5到1e-5，共25个，每个收敛标准收敛199次。一共11组共收敛了25*199*11次，对照组用重叠法训练网络共收敛了25*199*11次。因此这个实验共收敛了25*199*11*2次。

得到的实验数据

	平均准确率p-ave
δ	01	02	03	04	05	06	07	08	09	012	013
0.5	1.010562	0.985643	0.954759	0.972907	1.005327	1.010477	0.997612	0.972785	0.994202	1.057841	0.981816
0.4	0.95069	0.984626	0.970261	0.99017	0.984615	0.936324	0.941219	1.002714	0.9509	0.992104	0.909419
0.3	0.997544	0.978279	0.986308	0.988057	1.026692	0.989414	0.971975	1.01622	0.976514	0.969998	0.905215
0.2	0.997666	1.016997	1.003083	1.000811	1.013644	1.036889	0.955626	1.0228	0.996861	0.914024	0.93909
0.1	0.994973	0.998981	0.997497	1.004136	0.993289	1.049562	0.98977	1.006252	1.004545	0.98074	0.984664
0.01	0.998772	1.032332	0.99643	1.000166	0.997644	1.010251	0.996116	1.040274	0.999342	0.979411	0.999656
0.001	0.999106	0.993466	0.999629	0.999512	1.000155	1.002983	0.983877	0.994591	0.998418	0.995929	0.999194
9.00E-04	0.999343	0.995635	1	0.999246	1.000061	1.002021	0.988073	0.994341	0.994338	0.996676	1.000053
8.00E-04	0.999576	0.995988	1.001718	0.999398	0.999837	0.999973	0.993841	0.994025	0.990346	0.997103	1.000852
7.00E-04	1.000021	0.99852	1.003257	0.99975	1.000177	0.995462	0.994669	0.993162	0.98814	0.994989	1.000795
6.00E-04	0.999719	0.99896	0.998527	0.999662	1.000903	0.989662	0.993528	0.993871	0.987836	0.99486	0.999222
5.00E-04	0.999579	0.999729	0.983411	0.999739	1.000594	0.986574	0.993329	0.997045	0.988553	0.995047	0.998661
4.00E-04	0.999717	1.002456	0.979308	0.999812	0.998787	0.988348	0.994671	0.999903	0.991453	0.9928	1.000275
3.00E-04	0.998776	1.004612	0.989602	1.000826	0.996718	0.991392	0.997367	1.00447	0.99729	0.993128	0.999601
2.00E-04	0.998894	0.999062	0.995864	1.002085	0.996421	0.992696	0.997817	1.007148	0.993548	0.994696	0.996487
1.00E-04	1.000852	1.002931	1.002818	0.994955	1.006258	0.999963	0.999249	1.011294	0.993828	0.996692	0.999001
9.00E-05	1.000617	1.003625	1.003767	0.994636	1.0058	0.999979	0.998419	1.013065	0.992941	0.996719	1.000574
8.00E-05	1.000388	1.004641	1.003945	0.993841	1.005845	0.999674	0.997906	1.012799	0.990986	0.996571	1.000074
7.00E-05	1.000174	1.004352	1.003379	0.99521	1.005561	1.001128	0.996904	1.013894	0.989524	0.996029	0.999591
6.00E-05	0.999841	1.003942	1.002291	0.997387	1.006394	1.001567	0.996155	1.012944	0.990426	0.996065	0.999802
5.00E-05	0.99928	1.00316	1.000688	0.99984	1.008157	1.00489	0.99598	1.014733	0.992503	0.996005	0.999348
4.00E-05	0.999047	1.002764	1.001069	1.000356	1.011427	1.014654	0.996366	1.021431	0.993715	0.995526	0.998539
3.00E-05	0.999082	1.003092	1.004241	1.000532	1.011903	1.020132	0.99724	1.03468	0.994698	0.995441	0.997658
2.00E-05	0.999175	1.002069	1.005558	0.998564	1.012158	1.004678	0.999887	1.034621	0.995565	0.994653	0.995774
1.00E-05	0.999268	1.003453	1.004287	0.993217	1.009703	1.002874	1.003853	1.009891	0.999042	0.992808	0.99342

ave	0.999772	1.003403	1.003204	0.996854	1.008321	1.004954	0.998196	1.017935	0.993323	0.995651	0.998378

首先比较平均分类准确率，表格中的数据使用对应的重叠法的平均分类准确率/正常输入得到的平均分类准确率。

数据统计1e-5<=δ<=1e-4区间的平均值。网络02，03，05，06，08的比值都是显著大于1的，表明重叠法有效的提升了这些网络的性能。对这5个网络来说非重叠区域携带的信息只是干扰。重叠区域已经携带了决定分类的所有信息。

但是对于网络01，04，07，09，012，013，这6个网络用重叠法事实上导致了分类性能的下降，表明非重叠区与仍然有决定分类的重要信息，舍弃这部分信息导致了分类准确率的下降。

但是从绝对值来看性能下降的最大降幅也只有1-0.993323（网络09）用重叠法也仅仅使准确率下降了不到千分之7.这个实验至少说明了重叠区域携带了决定网络分类的绝大部分信息。

再来比较最大分类准确率

	最大值p-max
δ	01	02	03	04	05	06	07	08	09	012	013
0.5	0.960077	0.925291	0.833613	0.933662	0.930468	1.041563	0.935602	0.895297	0.961894	1.014138	0.883796
0.4	0.989573	0.991224	0.992795	1.001036	1.020443	1.022602	0.988241	1.009554	0.997409	0.987271	0.947027
0.3	0.999054	1	0.993289	1.005723	1.017544	1.026849	0.991269	1.017479	0.999481	0.990661	0.937434
0.2	0.999527	1.007228	1.00567	1.002585	1.011745	1.023991	0.991816	1.012638	1.001552	0.936228	0.961615
0.1	0.999527	1	1.002561	1.002579	0.99613	1.040952	0.997955	1.002088	1.004134	0.986586	0.996033
0.01	0.999053	0.998474	1.003592	1	0.998348	1.001058	0.996425	1.006296	0.999489	0.989839	1.001318
0.001	1.000473	0.997972	0.99949	1.001027	0.996721	0.992685	0.995927	0.997403	0.997457	0.999675	1.001956
9.00E-04	0.999527	0.997467	0.997964	1.001026	0.996177	0.993208	0.994399	0.996885	0.99695	0.999027	0.99935
8.00E-04	0.999527	0.997973	0.99695	1.001027	0.997268	0.993208	0.99389	0.997405	0.99695	0.997085	0.999675
7.00E-04	1	0.996962	0.99695	1.001026	0.997813	0.993208	0.993893	0.995339	0.995427	0.996437	0.99935
6.00E-04	1	0.99696	0.99695	1.000513	0.999452	0.992167	0.997454	0.995868	0.992378	0.997085	0.998052
5.00E-04	1	0.997975	0.996441	1	0.996723	0.98799	0.997454	0.995351	0.991874	0.995474	0.998051
4.00E-04	0.998582	0.996965	0.996441	0.999488	0.99509	0.990601	0.997454	0.998451	0.992886	0.995474	1.000325
3.00E-04	0.998582	0.996967	0.998475	0.998975	1.003268	0.990601	0.997964	0.998451	0.992378	0.995804	0.99935
2.00E-04	0.999054	1.001517	0.999492	1.000513	1.004897	0.993714	0.997964	1.008868	0.992378	0.998063	1
1.00E-04	1.000473	1.003537	1.002541	1.001026	1.006536	0.99843	0.997964	1.007792	0.992374	0.998066	1.001947
9.00E-05	1	1.001009	1.00305	1.001026	1.005988	0.998953	0.997457	1.007273	0.992883	0.998389	1.003892
8.00E-05	1	1.002524	1.002541	1.001026	1.005441	0.999477	0.997964	1.005711	0.991866	0.99839	1.003244
7.00E-05	1	1.003027	1.002033	1.00154	1.007077	1	0.997455	1.006231	0.992374	0.997102	1.002594
6.00E-05	1	1.003027	1.002031	1.002054	1.006532	1.000523	0.998472	1.007269	0.992374	0.997426	1.001296
5.00E-05	1	1.003027	1.003049	1.00154	1.005985	1.000523	1.002038	1.007269	0.993391	0.997103	0.999031
4.00E-05	0.999527	1.002521	1.004065	1	1.008719	1.000523	1.002546	1.005187	0.992374	0.997427	1.002264
3.00E-05	0.999527	1.002521	1.003555	0.999488	1.013691	1	1.002036	1.005708	0.993388	0.996465	0.998709
2.00E-05	0.999527	1.002521	1.005589	0.998975	1.010917	1.001569	1.001525	1.004143	0.996432	0.996789	0.998065
1.00E-05	0.999527	1.003026	1.005081	0.998976	1.009799	1.001045	1.002542	1.003102	0.995927	0.996149	0.998068

ave	0.999858	1.002674	1.003354	1.000565	1.008068	1.000104	1	1.005969	0.993338	0.997331	1.000911

这组数据同样使用对应的重叠法的数据/正常得到的数据，

网络02，03，05，06，08的比值仍然是大于1的，表明重叠法强化了网络的性能，也再次确认了这几个网络中重叠区域的重要价值。

网络04，013的比值也是大于1的，表明这两个网络的非重叠区域携带的信息的稀缺程度大于1/200，因为在约200次以内重叠区域携带的信息就可能覆盖因舍弃非重叠区域而丢弃的信息。由此07的非重叠区域信息的稀缺程度大约等于1/200，而01，09，012的非重叠部分携带的信息的稀缺程度要小于1/200，因为仅靠重叠部分在约200次以内也未能超过正常状态下产生的分类准确率最大值。

这组数据表明对于02，03，04，05，06，07，08，013这8个网络重叠区域有不小于1/200的概率携带了决定分类的全部有效信息。

而对于01，09，012这3个网络重叠区域携带全部有效信息的概率小于1/200.如果用收敛标准为1e-5<=δ<=1e-4区间的平均分类准确率的值来衡量，01，09，012这三个网络重叠部分携带的有效信息的损失至多不超过7‰，网络09，0.993338.

这组数据也侧面说明了重叠区域的价值。

综合这两组数字可以相对客观的得出结论，神经网络分类图片的重叠区域携带了决定分类的绝大部分信息，可以对图片分类产生决定性影响。

数据

正常	01	无核			正常	02	无核

迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max	迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max
7.949749	0.519843	0.5	0.069267	0.982979	9.356784	0.519446	0.5	0.066583	0.898111
151.7035	0.973848	0.4	0.074317	0.997636	210.9598	0.866532	0.4	0.078517	0.962724
194.2161	0.997384	0.3	0.07415	0.999527	269.7789	0.950133	0.3	0.081717	0.965706
234.8291	0.99765	0.2	0.077567	0.999527	325.6332	0.939239	0.2	0.083533	0.962724
310.402	0.99779	0.1	0.082033	0.999527	411.1156	0.958358	0.1	0.088883	0.968688
647.7839	0.998629	0.01	0.102883	0.999054	685.995	0.939326	0.01	0.108333	0.977137
2007.382	0.998734	0.001	0.18645	0.999054	1432.593	0.9767	0.001	0.157117	0.980119
2099.392	0.998684	9.00E-04	0.1907	0.999054	1469.94	0.976775	9.00E-04	0.1592	0.981113
2167.005	0.998539	8.00E-04	0.198683	0.999054	1474.804	0.97671	8.00E-04	0.156567	0.980616
2340.663	0.998175	7.00E-04	0.21225	0.999054	1541.688	0.977002	7.00E-04	0.163983	0.98161
2620.94	0.998501	6.00E-04	0.2223	0.999054	1747.91	0.977669	6.00E-04	0.17505	0.981113
2864.603	0.998719	5.00E-04	0.236067	0.999054	1953.216	0.977677	5.00E-04	0.18705	0.98161
3077.377	0.99847	4.00E-04	0.251017	1	2189.528	0.975286	4.00E-04	0.202517	0.982604
4194.266	0.999356	3.00E-04	0.322033	1	2351.286	0.973735	3.00E-04	0.21545	0.983101
5112.06	0.999354	2.00E-04	0.376917	1	2728.874	0.97986	2.00E-04	0.235283	0.983101
5308.844	0.997814	1.00E-04	0.386233	0.999054	3229.085	0.980764	1.00E-04	0.269567	0.983598
5502.422	0.998094	9.00E-05	0.403217	0.999527	3394.925	0.980546	9.00E-05	0.27885	0.985586
5762.844	0.998413	8.00E-05	0.420183	0.999527	3563.246	0.980419	8.00E-05	0.294517	0.984592
6276.281	0.998655	7.00E-05	0.44645	0.999527	3710.432	0.980749	7.00E-05	0.299567	0.985089
7741.688	0.998981	6.00E-05	0.530517	0.999527	3956.523	0.981418	6.00E-05	0.315167	0.985089
9718.231	0.999439	5.00E-05	0.6587	0.999527	4213.608	0.982375	5.00E-05	0.33155	0.985089
11940.51	0.999527	4.00E-05	0.79265	0.999527	4531.337	0.982959	4.00E-05	0.352717	0.985586
14787.96	0.999499	3.00E-05	0.962917	0.999527	5093.397	0.983049	3.00E-05	0.3886	0.985586
15441.05	0.999389	2.00E-05	1.008633	0.999527	6601.477	0.983661	2.00E-05	0.488517	0.985586
23828.85	0.999347	1.00E-05	1.510667	0.999527	8579.095	0.982357	1.00E-05	0.6114	0.985586


重叠	01	无核			重叠	02	无核

迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max	迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max
12.02513	0.525334	0.5	0.06665	0.943735	9.281407	0.511988	0.5	0.068383	0.831014
280.7186	0.925828	0.4	0.080833	0.987234	340.2663	0.85321	0.4	0.08545	0.954274
360.3065	0.994934	0.3	0.084767	0.998582	426.0302	0.929496	0.3	0.090717	0.965706
411.8291	0.995322	0.2	0.090167	0.999054	500.1859	0.955203	0.2	0.095467	0.969682
530.4573	0.992775	0.1	0.094883	0.999054	644.2814	0.957381	0.1	0.104717	0.968688
893.6734	0.997403	0.01	0.1195	0.998109	1112.03	0.969697	0.01	0.141617	0.975646
1688.402	0.99784	0.001	0.177383	0.999527	2281.769	0.970319	0.001	0.212783	0.978131
1748.658	0.998028	9.00E-04	0.1786	0.998582	2366.276	0.972512	9.00E-04	0.216083	0.978628
1797.573	0.998116	8.00E-04	0.1737	0.998582	2404.467	0.972791	8.00E-04	0.217217	0.978628
1838.402	0.998197	7.00E-04	0.17575	0.999054	2597.276	0.975556	7.00E-04	0.230667	0.978628
1918.678	0.99822	6.00E-04	0.180867	0.999054	2865.508	0.976653	6.00E-04	0.247083	0.978131
2073.307	0.998299	5.00E-04	0.189333	0.999054	3190.231	0.977412	5.00E-04	0.268467	0.979622
2130.211	0.998187	4.00E-04	0.194433	0.998582	3387.332	0.977682	4.00E-04	0.280367	0.979622
2202.367	0.998133	3.00E-04	0.198833	0.998582	3880.598	0.978226	3.00E-04	0.31655	0.980119
2542.714	0.998249	2.00E-04	0.21915	0.999054	6708.151	0.97894	2.00E-04	0.4951	0.984592
3106.513	0.998665	1.00E-04	0.25245	0.999527	16564.7	0.983638	1.00E-04	1.128517	0.987078
3206.643	0.99871	9.00E-05	0.259583	0.999527	17734.32	0.9841	9.00E-05	1.205867	0.986581
3384.03	0.9988	8.00E-05	0.2703	0.999527	19353.75	0.98497	8.00E-05	1.3086	0.987078
3767.598	0.998829	7.00E-05	0.292033	0.999527	20519.56	0.985017	7.00E-05	1.383283	0.988072
4019.492	0.998822	6.00E-05	0.309367	0.999527	21291.9	0.985287	6.00E-05	1.429217	0.988072
4679.06	0.998719	5.00E-05	0.3517	0.999527	22457.05	0.985479	5.00E-05	1.5048	0.988072
5194.196	0.998574	4.00E-05	0.3824	0.999054	24930.77	0.985676	4.00E-05	1.668233	0.988072
5537.467	0.998582	3.00E-05	0.404433	0.999054	27828.73	0.986088	3.00E-05	1.8504	0.988072
6756.543	0.998565	2.00E-05	0.483933	0.999054	31770.94	0.985696	2.00E-05	2.034867	0.988072
8724.392	0.998615	1.00E-05	0.603817	0.999054	39501.2	0.985749	1.00E-05	2.475483	0.988569

正常	03	无核			正常	04	无核

迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max	迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max
10.03015	0.535721	0.5	0.064833	0.896985	8.703518	0.531851	0.5	0.06105	0.929664
214	0.88045	0.4	0.075033	0.976382	193.0402	0.900978	0.4	0.06925	0.9842
277.2714	0.953127	0.3	0.0773	0.973367	244.9347	0.968049	0.3	0.072283	0.979613
336.5477	0.966463	0.2	0.088583	0.974874	300.8492	0.979469	0.2	0.075667	0.985729
408.9447	0.972405	0.1	0.086733	0.980905	397.2513	0.98265	0.1	0.081883	0.988277
697.5678	0.969753	0.01	0.1023	0.979397	668.794	0.989904	0.01	0.098167	0.991845
1432.874	0.981334	0.001	0.150867	0.98593	1578.874	0.991102	0.001	0.155133	0.992864
1462.05	0.981336	9.00E-04	0.150983	0.987437	1687.93	0.991586	9.00E-04	0.15975	0.993374
1512.905	0.980273	8.00E-04	0.154217	0.988442	1786.176	0.992022	8.00E-04	0.164333	0.992864
1634.769	0.97624	7.00E-04	0.160517	0.988442	1879.774	0.992245	7.00E-04	0.171033	0.993374
1807.854	0.976872	6.00E-04	0.170817	0.988442	1986.151	0.992462	6.00E-04	0.176583	0.993884
2000.543	0.982435	5.00E-04	0.182967	0.988442	2082.462	0.99267	5.00E-04	0.181783	0.994393
2190.181	0.986298	4.00E-04	0.19675	0.988442	2180.704	0.993005	4.00E-04	0.187283	0.994903
2519.387	0.986478	3.00E-04	0.21445	0.988442	2511.181	0.992529	3.00E-04	0.2079	0.994903
3092.035	0.986556	2.00E-04	0.256717	0.988945	3000.452	0.991225	2.00E-04	0.236883	0.993374
4020.884	0.984798	1.00E-04	0.309267	0.988945	4260.503	0.99144	1.00E-04	0.312183	0.993374
4122.94	0.984766	9.00E-05	0.313367	0.988442	4498.633	0.991207	9.00E-05	0.326183	0.993374
4381.709	0.985036	8.00E-05	0.334133	0.988945	4747.538	0.990959	8.00E-05	0.340833	0.993374
4862.809	0.985604	7.00E-05	0.363017	0.988945	4898.653	0.990905	7.00E-05	0.3504	0.992864
5464.472	0.986374	6.00E-05	0.39855	0.989447	5077.508	0.991008	6.00E-05	0.361017	0.992355
6367.533	0.987508	5.00E-05	0.45765	0.988945	5352.915	0.991264	5.00E-05	0.378667	0.992864
7179.533	0.987061	4.00E-05	0.504817	0.988945	5996.151	0.991576	4.00E-05	0.416017	0.994393
8391	0.98547	3.00E-05	0.579817	0.989447	6741.417	0.99099	3.00E-05	0.463	0.994903
10055.13	0.984998	2.00E-05	0.685467	0.988945	8241.487	0.990111	2.00E-05	0.554067	0.994903
16955.01	0.985437	1.00E-05	1.129667	0.988945	12059.31	0.99341	1.00E-05	0.780033	0.995413


重叠	03	无核			重叠	04	无核

迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max	迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max
8.889447	0.511485	0.5	0.062833	0.747739	8.286432	0.517442	0.5	0.063067	0.867992
360.0101	0.854266	0.4	0.081217	0.969347	316.3367	0.892121	0.4	0.0788	0.985219
442.0603	0.940077	0.3	0.088633	0.966834	412.1759	0.956487	0.3	0.083967	0.985219
527.8492	0.969443	0.2	0.092967	0.980402	481.3216	0.980263	0.2	0.088217	0.988277
640.201	0.96997	0.1	0.09905	0.983417	596.201	0.986715	0.1	0.095883	0.990826
1027.613	0.966291	0.01	0.123783	0.982915	913.3668	0.990068	0.01	0.11455	0.991845
1882.196	0.98097	0.001	0.17665	0.985427	1502.462	0.990618	0.001	0.15395	0.993884
2124.603	0.981336	9.00E-04	0.191317	0.985427	1532.965	0.990838	9.00E-04	0.15475	0.994393
2310.317	0.981958	8.00E-04	0.203933	0.985427	1595.799	0.991425	8.00E-04	0.158283	0.993884
2434.894	0.97942	7.00E-04	0.2123	0.985427	1735.638	0.991996	7.00E-04	0.166033	0.994393
2514.709	0.975432	6.00E-04	0.214483	0.985427	1893.186	0.992127	6.00E-04	0.177	0.994393
2584.161	0.966137	5.00E-04	0.219783	0.984925	2071.789	0.992411	5.00E-04	0.1873	0.994393
2788.864	0.96589	4.00E-04	0.232033	0.984925	2195.186	0.992818	4.00E-04	0.19595	0.994393
3581.623	0.97622	3.00E-04	0.281383	0.986935	2344.08	0.993348	3.00E-04	0.20405	0.993884
7254.774	0.982475	2.00E-04	0.511	0.988442	2361.397	0.993292	2.00E-04	0.205967	0.993884
15178.34	0.987574	1.00E-04	0.995767	0.991457	2617.96	0.986438	1.00E-04	0.221767	0.994393
16403	0.988475	9.00E-05	1.071233	0.991457	2701.548	0.98589	9.00E-05	0.226383	0.994393
17599.95	0.988922	8.00E-05	1.142583	0.991457	2766.181	0.984855	8.00E-05	0.23305	0.994393
18170.47	0.988935	7.00E-05	1.178567	0.990955	3034.291	0.986159	7.00E-05	0.242183	0.994393
18399.82	0.988634	6.00E-05	1.215167	0.991457	3521.397	0.988418	6.00E-05	0.26965	0.994393
19024.32	0.988187	5.00E-05	1.2372	0.99196	4362.503	0.991105	5.00E-05	0.320283	0.994393
20923.2	0.988116	4.00E-05	1.346367	0.992965	6561.136	0.99193	4.00E-05	0.4582	0.994393
24711.29	0.989649	3.00E-05	1.5876	0.992965	11045.08	0.991517	3.00E-05	0.745217	0.994393
28002.55	0.990472	2.00E-05	1.788917	0.994472	12777.64	0.98869	2.00E-05	0.8512	0.993884
34822.22	0.989662	1.00E-05	2.199683	0.99397	20794.52	0.986671	1.00E-05	1.3479	0.994393

正常	05	无核			正常	06	无核

迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max	迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max
8.849246	0.520068	0.5	0.059417	0.845085	8.386935	0.522183	0.5	0.061417	0.831785
258.8844	0.816929	0.4	0.070267	0.940705	214.6884	0.812458	0.4	0.07035	0.936017
347.7588	0.904195	0.3	0.07575	0.94391	272.2462	0.910416	0.3	0.073633	0.941692
440.6784	0.937473	0.2	0.0812	0.955128	326.4121	0.926786	0.2	0.077267	0.946336
547.8794	0.962038	0.1	0.0878	0.966346	392.809	0.916919	0.1	0.081967	0.932405
854.9296	0.966	0.01	0.108267	0.970085	725.2663	0.962503	0.01	0.100733	0.975748
1520.281	0.972582	0.001	0.145783	0.977564	1655.633	0.974454	0.001	0.15855	0.987616
1565.859	0.972751	9.00E-04	0.148717	0.978098	1715.487	0.974859	9.00E-04	0.161117	0.987616
1663.829	0.972998	8.00E-04	0.155017	0.977564	1746.543	0.976036	8.00E-04	0.16255	0.987616
1722.08	0.972727	7.00E-04	0.158933	0.97703	1809.07	0.978751	7.00E-04	0.1662	0.987616
1796.01	0.971975	6.00E-04	0.162933	0.975427	1917.171	0.982936	6.00E-04	0.172933	0.988132
1860.814	0.971613	5.00E-04	0.165967	0.978098	2015.337	0.9859	5.00E-04	0.17845	0.988132
1976.392	0.971876	4.00E-04	0.17305	0.979167	2152.101	0.984279	4.00E-04	0.187067	0.988132
2400.894	0.974193	3.00E-04	0.19965	0.980769	2468.668	0.981385	3.00E-04	0.205667	0.988132
2889.005	0.978804	2.00E-04	0.22725	0.981838	3020.688	0.981647	2.00E-04	0.238683	0.985036
3654.734	0.978898	1.00E-04	0.272383	0.980769	4026.884	0.978466	1.00E-04	0.298833	0.986068
3803.186	0.978904	9.00E-05	0.280833	0.981303	4574.678	0.978847	9.00E-05	0.33145	0.986068
4038.543	0.978646	8.00E-05	0.295633	0.981838	5040.045	0.979822	8.00E-05	0.359617	0.986068
4262.251	0.978882	7.00E-05	0.309167	0.981303	5518.402	0.979321	7.00E-05	0.3878	0.986068
4393.869	0.978254	6.00E-05	0.315733	0.981303	5989.905	0.979337	6.00E-05	0.416083	0.986068
4525.648	0.976756	5.00E-05	0.324	0.981838	6524.844	0.976806	5.00E-05	0.4516	0.986068
4641.809	0.97441	4.00E-05	0.33075	0.980235	7490.151	0.968387	4.00E-05	0.506683	0.986068
4650	0.973999	3.00E-05	0.331533	0.975427	9533.332	0.964547	3.00E-05	0.62935	0.986068
4662.191	0.97415	2.00E-05	0.332783	0.978632	12042.67	0.980001	2.00E-05	0.781567	0.986584
5357.417	0.975803	1.00E-05	0.374083	0.981303	21193.02	0.982607	1.00E-05	1.330567	0.9871


重叠	05	无核			重叠	06	无核

迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max	迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max
11.38191	0.522839	0.5	0.061767	0.786325	9.38191	0.527654	0.5	0.06195	0.866357
500.4171	0.804361	0.4	0.088317	0.959936	360.2663	0.760723	0.4	0.079933	0.957172
634.3367	0.92833	0.3	0.094383	0.96047	464.0603	0.900778	0.3	0.086667	0.966976
736.7638	0.950264	0.2	0.10265	0.966346	551.4673	0.960974	0.2	0.092517	0.96904
925.196	0.955582	0.1	0.11305	0.962607	628.2663	0.962363	0.1	0.097333	0.970588
1551.663	0.963724	0.01	0.155867	0.968483	1134.523	0.97237	0.01	0.132517	0.97678
2944	0.972732	0.001	0.2403	0.974359	2117.955	0.977361	0.001	0.189717	0.980392
2944	0.97281	9.00E-04	0.24015	0.974359	2209.317	0.976829	9.00E-04	0.1963	0.980908
2944	0.97284	8.00E-04	0.237183	0.974893	2285.407	0.97601	8.00E-04	0.200517	0.980908
2963.819	0.972899	7.00E-04	0.238867	0.974893	2399.417	0.974309	7.00E-04	0.207583	0.980908
2950.683	0.972853	6.00E-04	0.239367	0.974893	2474.271	0.972774	6.00E-04	0.21095	0.980392
3064.844	0.97219	5.00E-04	0.245217	0.974893	2559.719	0.972663	5.00E-04	0.2164	0.976264
3313.186	0.970698	4.00E-04	0.263183	0.974359	2884.945	0.97281	4.00E-04	0.235783	0.978844
4489.688	0.970995	3.00E-04	0.335467	0.983974	3903.482	0.972937	3.00E-04	0.300067	0.978844
10070.31	0.975301	2.00E-04	0.68315	0.986645	5471.648	0.974478	2.00E-04	0.402083	0.978844
18300.75	0.985024	1.00E-04	1.194817	0.987179	11467.41	0.978429	1.00E-04	0.768733	0.98452
18760.43	0.984581	9.00E-05	1.22635	0.987179	14578.9	0.978826	9.00E-05	0.9608	0.985036
18960.83	0.984366	8.00E-05	1.235833	0.987179	18135.92	0.979503	8.00E-05	1.213633	0.985552
19897.09	0.984326	7.00E-05	1.293483	0.988248	20685.69	0.980426	7.00E-05	1.33825	0.986068
20744.52	0.984509	6.00E-05	1.3515	0.987714	23503.52	0.980872	6.00E-05	1.557783	0.986584
22494.59	0.984723	5.00E-05	1.45665	0.987714	26049.64	0.981582	5.00E-05	1.69695	0.986584
25776.8	0.985545	4.00E-05	1.6611	0.988782	28677.2	0.982578	4.00E-05	1.8414	0.986584
28870.21	0.985593	3.00E-05	1.856183	0.988782	32251.8	0.983965	3.00E-05	2.036517	0.986068
36250.91	0.985993	2.00E-05	2.317567	0.989316	36272.02	0.984585	2.00E-05	2.52395	0.988132
50208.46	0.985271	1.00E-05	3.185233	0.990919	44806.67	0.98543	1.00E-05	3.107433	0.988132

正常	07	无核			正常	08	无核

迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max	迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max
9.934673	0.534671	0.5	0.061217	0.964646	9.075377	0.529462	0.5	0.061233	0.914023
174.3769	0.938351	0.4	0.068967	0.987879	212.4221	0.890779	0.4	0.069617	0.964176
220.8593	0.973786	0.3	0.071567	0.983333	268.4673	0.948591	0.3	0.07385	0.966223
262.5879	0.981237	0.2	0.073717	0.987374	330.9347	0.95752	0.2	0.0773	0.971853
358.5628	0.985384	0.1	0.079667	0.987879	417.7337	0.96917	0.1	0.082533	0.980553
644.9246	0.986671	0.01	0.096667	0.988889	695.7337	0.93887	0.01	0.098717	0.975435
1388.201	0.983635	0.001	0.1463	0.991919	1758.704	0.971904	0.001	0.166883	0.985159
1420.211	0.982838	9.00E-04	0.1425	0.991919	1781.095	0.972081	9.00E-04	0.1637	0.98567
1451.477	0.983562	8.00E-04	0.1449	0.991919	1801.231	0.972251	8.00E-04	0.166383	0.986182
1568.668	0.987455	7.00E-04	0.157033	0.992424	1894.121	0.974885	7.00E-04	0.171633	0.988229
1733.327	0.990516	6.00E-04	0.161817	0.991919	2070.497	0.978534	6.00E-04	0.1826	0.990788
1876.704	0.99146	5.00E-04	0.17385	0.991919	2299.965	0.981553	5.00E-04	0.197417	0.990788
1972.804	0.991013	4.00E-04	0.179417	0.991919	2553.03	0.982669	4.00E-04	0.213683	0.9913
2032.754	0.990973	3.00E-04	0.180417	0.992424	2841.08	0.980962	3.00E-04	0.228767	0.9913
2321.186	0.99147	2.00E-04	0.199217	0.992424	3064.146	0.979313	2.00E-04	0.24365	0.981064
3413.477	0.98996	1.00E-04	0.263217	0.992424	3758.658	0.977966	1.00E-04	0.286083	0.985159
3712.613	0.990231	9.00E-05	0.282417	0.992929	3942.387	0.976731	9.00E-05	0.298	0.985159
3966.492	0.990325	8.00E-05	0.298067	0.992424	4181.824	0.976952	8.00E-05	0.311933	0.98567
4163.055	0.99029	7.00E-05	0.3118	0.992424	4468.92	0.976369	7.00E-05	0.328867	0.98567
4198.864	0.990138	6.00E-05	0.313183	0.991414	5123.693	0.977737	6.00E-05	0.373883	0.98567
4251.136	0.989828	5.00E-05	0.314183	0.991414	6083.332	0.975787	5.00E-05	0.426967	0.98567
4384.543	0.989577	4.00E-05	0.3258	0.991919	7145.613	0.969116	4.00E-05	0.490733	0.986694
4661.377	0.989582	3.00E-05	0.344617	0.992424	9593.352	0.956525	3.00E-05	0.644333	0.986182
5864.201	0.991328	2.00E-05	0.41705	0.993434	11486.03	0.956592	2.00E-05	0.74435	0.988229
8022.01	0.989924	1.00E-05	0.546867	0.993434	23072.07	0.980779	1.00E-05	1.4413	0.989765


重叠	07	无核			重叠	08	无核

迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max	迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max
10.34673	0.533394	0.5	0.066517	0.902525	9.301508	0.515052	0.5	0.065967	0.818321
297.9698	0.883194	0.4	0.083017	0.976263	392.7437	0.893197	0.4	0.0858	0.973388
401.9196	0.946495	0.3	0.09155	0.974747	509.0101	0.963978	0.3	0.090883	0.983112
459.4472	0.937696	0.2	0.093333	0.979293	582.1307	0.979352	0.2	0.096383	0.984135
586.2613	0.975303	0.1	0.10175	0.985859	662.402	0.975229	0.1	0.1014	0.9826
943.1156	0.982838	0.01	0.125183	0.985354	995.5779	0.976682	0.01	0.1281	0.981576
1721.261	0.967776	0.001	0.184833	0.987879	2183.266	0.966647	0.001	0.20225	0.9826
1788.583	0.971116	9.00E-04	0.182667	0.986364	2202.884	0.966581	9.00E-04	0.20605	0.9826
1873.618	0.977504	8.00E-04	0.18845	0.985859	2192.136	0.966442	8.00E-04	0.202317	0.983623
1970.236	0.982191	7.00E-04	0.195633	0.986364	2298.045	0.968219	7.00E-04	0.210883	0.983623
2011.729	0.984105	6.00E-04	0.197533	0.989394	2617.784	0.972537	6.00E-04	0.232883	0.986694
2036.095	0.984846	5.00E-04	0.2013	0.989394	3198.065	0.978652	5.00E-04	0.2693	0.986182
2074.201	0.985732	4.00E-04	0.203	0.989394	4470.07	0.982574	4.00E-04	0.365333	0.989765
2189.94	0.988364	3.00E-04	0.211	0.990404	6871.824	0.985346	3.00E-04	0.5771	0.989765
2742.543	0.989305	2.00E-04	0.253033	0.990404	12043.54	0.986313	2.00E-04	0.962317	0.989765
4219.487	0.989216	1.00E-04	0.3298	0.990404	15740.55	0.989011	1.00E-04	1.1409	0.992835
4442.905	0.988666	9.00E-05	0.343867	0.990404	16066.14	0.989492	9.00E-05	1.163167	0.992323
4679.628	0.988252	8.00E-05	0.358283	0.990404	17083.9	0.989456	8.00E-05	1.2021	0.9913
5149.196	0.987224	7.00E-05	0.3836	0.989899	18034.4	0.989934	7.00E-05	1.258217	0.991812
5628.241	0.986331	6.00E-05	0.417467	0.989899	19701.26	0.990392	6.00E-05	1.348833	0.992835
5871.035	0.985848	5.00E-05	0.445733	0.993434	21148.4	0.990163	5.00E-05	1.428267	0.992835
6272.633	0.98598	4.00E-05	0.483917	0.994444	22817.89	0.989885	4.00E-05	1.535283	0.991812
8910.432	0.986851	3.00E-05	0.661333	0.994444	25249.14	0.989698	3.00E-05	1.653283	0.991812
21818.84	0.991216	2.00E-05	1.5182	0.994949	31225.81	0.989711	2.00E-05	2.007817	0.992323
37734.63	0.993739	1.00E-05	2.546983	0.99596	47133.26	0.99048	1.00E-05	2.89545	0.992835

正常	09	无核			正常	012	无核

迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max	迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max
8.59799	0.5312	0.5	0.06205	0.92358	143.0804	0.474752	0.5	0.109017	0.831586
177.6533	0.905556	0.4	0.069867	0.970337	492.4372	0.768878	0.4	0.12575	0.848745
235.196	0.959102	0.3	0.073133	0.968829	592.0603	0.807011	0.3	0.132233	0.884652
286.4372	0.967424	0.2	0.076117	0.971845	915.3065	0.913217	0.2	0.152633	0.936765
374.5126	0.969346	0.1	0.082833	0.972851	1148.432	0.926313	0.1	0.16685	0.947569
702.0854	0.979301	0.01	0.105967	0.984414	2368.462	0.963258	0.01	0.245533	0.969495
1824.538	0.983482	0.001	0.181567	0.988436	7079.497	0.973415	0.001	0.532883	0.97871
1923.593	0.985733	9.00E-04	0.175083	0.988939	7342.136	0.971767	9.00E-04	0.547817	0.979981
1983.005	0.987143	8.00E-04	0.177783	0.988939	7735.859	0.971111	8.00E-04	0.57255	0.981252
2011.397	0.98733	7.00E-04	0.180833	0.989442	8218.824	0.973777	7.00E-04	0.600783	0.980934
2064.101	0.987158	6.00E-04	0.1831	0.989442	9161.719	0.97649	6.00E-04	0.6589	0.981252
2142.317	0.986539	5.00E-04	0.188467	0.989945	10863.08	0.977514	5.00E-04	0.767467	0.982841
2299.291	0.984627	4.00E-04	0.198083	0.989442	12474.05	0.980008	4.00E-04	0.866183	0.982841
2657.523	0.982565	3.00E-04	0.221633	0.989442	13457.31	0.980973	3.00E-04	0.940017	0.98443
3488.905	0.986729	2.00E-04	0.277183	0.989442	17356.45	0.981999	2.00E-04	1.156483	0.984112
4566.955	0.982077	1.00E-04	0.338317	0.988939	27869.65	0.982799	1.00E-04	1.859983	0.986018
4747.884	0.98284	9.00E-05	0.3475	0.988939	29662.53	0.983055	9.00E-05	1.950983	0.986336
5123.704	0.984899	8.00E-05	0.373883	0.988939	31924.66	0.983594	8.00E-05	2.0826	0.986654
5633.075	0.987115	7.00E-05	0.4039	0.988939	35684.52	0.984061	7.00E-05	2.332867	0.986972
5960.563	0.987433	6.00E-05	0.417633	0.988939	39541.35	0.984366	6.00E-05	2.55965	0.987607
6387.477	0.986713	5.00E-05	0.443917	0.988939	44577.43	0.984805	5.00E-05	2.8795	0.987289
6903.236	0.986062	4.00E-05	0.47415	0.988939	51671.21	0.985383	4.00E-05	3.336733	0.987925
7474.347	0.985377	3.00E-05	0.50815	0.988436	59172.44	0.985943	3.00E-05	3.779117	0.988878
8212.025	0.984429	2.00E-05	0.551583	0.986425	75528.11	0.986898	2.00E-05	4.76705	0.989514
12200.83	0.981115	1.00E-05	0.792583	0.987431	105716.9	0.988208	1.00E-05	6.640533	0.990149


重叠	09	无核			重叠	012	无核

迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max	迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max
10.51759	0.528121	0.5	0.062683	0.888386	374.0201	0.502212	0.5	0.124033	0.843343
307.2864	0.861093	0.4	0.0773	0.967823	1011.196	0.762807	0.4	0.154417	0.837941
402.9397	0.936576	0.3	0.083083	0.968326	1189.06	0.782799	0.3	0.166367	0.87639
483.7789	0.964387	0.2	0.0885	0.973353	1549.889	0.834703	0.2	0.185117	0.877026
599.3417	0.973753	0.1	0.095167	0.976873	2424.653	0.908471	0.1	0.236567	0.934859
1027.156	0.978656	0.01	0.124017	0.983912	6424.362	0.943425	0.01	0.467867	0.959644
2130.734	0.981926	0.001	0.18805	0.985923	30153.91	0.969452	0.001	1.846983	0.978392
2202.412	0.980152	9.00E-04	0.192633	0.985923	30740.28	0.968537	9.00E-04	1.885367	0.979028
2283.065	0.977613	8.00E-04	0.194017	0.985923	33054.72	0.968297	8.00E-04	2.050567	0.978392
2339.497	0.97562	7.00E-04	0.201867	0.984917	34977.72	0.968898	7.00E-04	2.15735	0.977439
2357.166	0.97515	6.00E-04	0.20465	0.9819	40055.82	0.971472	6.00E-04	2.45425	0.978392
2363.709	0.975246	5.00E-04	0.2042	0.9819	44602.76	0.972672	5.00E-04	2.732417	0.978392
2460.452	0.976211	4.00E-04	0.2098	0.982403	51495.46	0.972952	4.00E-04	3.1383	0.978392
2927.035	0.979902	3.00E-04	0.23475	0.9819	64812.85	0.974231	3.00E-04	4.053317	0.980299
3496.05	0.980362	2.00E-04	0.273617	0.9819	90092.04	0.976791	2.00E-04	5.641317	0.982205
3851.709	0.976016	1.00E-04	0.293733	0.981398	151003.7	0.979548	1.00E-04	9.345617	0.984112
3904.945	0.975903	9.00E-05	0.29505	0.9819	154713	0.979829	9.00E-05	9.618083	0.984747
4339.859	0.976021	8.00E-05	0.3248	0.980895	167554.5	0.980222	8.00E-05	10.25902	0.985065
5864.171	0.976774	7.00E-05	0.41625	0.981398	181154.7	0.980153	7.00E-05	10.51042	0.984112
8626.462	0.977979	6.00E-05	0.586567	0.981398	196250.7	0.980492	6.00E-05	11.26847	0.985065
11674.61	0.979316	5.00E-05	0.7653	0.982403	228451.7	0.98087	5.00E-05	13.62085	0.98443
13097.46	0.979864	4.00E-05	0.85945	0.981398	260914.6	0.980974	4.00E-05	15.60588	0.985383
15809.87	0.980152	3.00E-05	1.019833	0.9819	304038.9	0.981448	3.00E-05	18.24828	0.985383
21283.45	0.980064	2.00E-05	1.3524	0.982906	391467.8	0.981621	2.00E-05	24.75012	0.986336
26430.05	0.980175	1.00E-05	1.663517	0.983409	563334.6	0.9811	1.00E-05	33.42447	0.986336

正常	013	无核

迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max
182.0151	0.514505	0.5	0.10805	0.88672
498.8693	0.797229	0.4	0.125367	0.888
587.3668	0.851805	0.3	0.129833	0.9104
838.3467	0.935685	0.2	0.1466	0.95872
1157.206	0.956699	0.1	0.165017	0.968
2262.548	0.968476	0.01	0.231783	0.97152
5821.432	0.979145	0.001	0.4492	0.98176
6093.719	0.979549	9.00E-04	0.46375	0.98432
6433.141	0.979592	8.00E-04	0.487483	0.984
7295.126	0.979412	7.00E-04	0.538017	0.98432
9262.221	0.979797	6.00E-04	0.656033	0.9856
11339.58	0.978706	5.00E-04	0.778433	0.98496
11943.81	0.97705	4.00E-04	0.818167	0.984
14694.03	0.979666	3.00E-04	0.98225	0.98496
17592.66	0.981914	2.00E-04	1.15855	0.98464
25758.62	0.980649	1.00E-04	1.659633	0.98592
26923.39	0.980814	9.00E-05	1.722633	0.98656
30040.58	0.981213	8.00E-05	1.915633	0.98656
33474.66	0.982304	7.00E-05	2.119333	0.98688
38849.34	0.98299	6.00E-05	2.479933	0.98784
49878.19	0.98381	5.00E-05	3.117667	0.99072
59549.78	0.98514	4.00E-05	3.707583	0.98944
77231.55	0.987306	3.00E-05	4.76895	0.99168
96534.81	0.98943	2.00E-05	6.0017	0.992
120712.4	0.99123	1.00E-05	7.573317	0.9936



重叠	013	无核

迭代次数n	平均准确率p-ave	δ	耗时 min/199	最大值p-max
404.9799	0.50515	0.5	0.12655	0.78368
991.0603	0.725015	0.4	0.15885	0.84096
1238.271	0.771067	0.3	0.171683	0.85344
1825.693	0.878693	0.2	0.198617	0.92192
2581.794	0.942027	0.1	0.239783	0.96416
5755.588	0.968143	0.01	0.424167	0.9728
30075.48	0.978356	0.001	1.853917	0.98368
31605.11	0.9796	9.00E-04	1.918483	0.98368
33099.97	0.980427	8.00E-04	2.013283	0.98368
34928.8	0.980191	7.00E-04	2.112617	0.98368
36477.69	0.979034	6.00E-04	2.190933	0.98368
38330.22	0.977396	5.00E-04	2.289883	0.98304
41765.76	0.977319	4.00E-04	2.494833	0.98432
48434.08	0.979276	3.00E-04	2.8846	0.98432
59792.44	0.978465	2.00E-04	3.501383	0.98464
106674.1	0.97967	1.00E-04	6.115683	0.98784
127594.4	0.981377	9.00E-05	7.3852	0.9904
142719.9	0.981286	8.00E-05	8.214883	0.98976
167289.7	0.981902	7.00E-05	9.658817	0.98944
186063	0.982796	6.00E-05	10.58245	0.98912
222370.8	0.983169	5.00E-05	12.74997	0.98976
262578.8	0.983701	4.00E-05	15.24528	0.99168
337028.5	0.984994	3.00E-05	19.39107	0.9904
427944.8	0.985249	2.00E-05	24.07158	0.99008
688638.7	0.984708	1.00E-05	38.47677	0.99168

黑榆

关注

0
点赞
踩
0

收藏

觉得还不错? 一键收藏
打赏
0
评论
实验：是否图片的重叠区域携带了决定分类的所有信息？

使用重叠法将参与分类训练的同一批次图片的所有不重叠部分变成0，只保留图片的重叠部分训练网络，如果分类准确率上升表明重叠部分已经包含决定分类的全部信息。如果这个假设成立，表明神经网络实现分类是通过在分类对象之间建立一个公共的重叠区，并通过识别重叠区来实现的。具体实验过程 mnist A B C ...
复制链接

扫一扫