刑事案件现场图自动分类算法 下载: 1147次
Automated Classification Method for Crime Scene Sketches
中国人民公安大学侦查与刑事科学技术学院, 北京 100038
图 & 表
图 1. 规范现场图。(a)现场平面示意图;(b)自绘方位示意图;(c)地图方位示意图
Fig. 1. Specification of crime scene sketches. (a) Crime scene overview sketch; (b) self-drawn location sketch; (c) map-screenshot location sketch
下载图片 查看原文
图 2. XCTNet结构图
Fig. 2. Diagram of XCTNet structure
下载图片 查看原文
图 3. Inception结构图
Fig. 3. Structure of Inception
下载图片 查看原文
图 4. 准确率和ROC曲线。(a)不同网络结构的准确率对比;(b) ROC曲线
Fig. 4. Accuracy and ROC curves. (a) Comparison of accuracies of different network structures; (b) ROC curves
下载图片 查看原文
图 5. 分类错误图片
Fig. 5. Misclassification pictures
下载图片 查看原文
表 1Inception构成参数
Table1. Composition parameters of Inception
Structure | conv1_1 | conv2_1 | conv2_2 | conv3_1 | conv3_2 | conv4_2 | Output |
---|
Inception1a | 64 | 96 | 128 | 16 | 32 | 32 | 256 | Inception1b | 128 | 128 | 192 | 32 | 96 | 64 | 480 | Inception2a | 256 | 160 | 320 | 32 | 128 | 128 | 832 | Inception2b | 384 | 192 | 384 | 48 | 128 | 128 | 1024 |
|
查看原文
表 2数据集组成
Table2. Composition of dataset
| Training dataset | Validation dataset | Test dataset |
---|
Crime dataset scene overview sketch | 17101 | 2138 | 2137 | Self-drawn location sketch | 10005 | 1251 | 1250 | Map-screenshot location sketch | 24173 | 3022 | 3021 | Photograph | 21730 | 2716 | 2716 | Total | 73009 | 9127 | 9124 |
|
查看原文
表 3检测精度与参数
Table3. Detection accuracy and parameters
| Accuracy on testset /% | Parameter quantity |
---|
AlexNet | 94.87 | 56884612 | XCTNet | 98.65 | 4935460 |
|
查看原文
表 4混淆矩阵
Table4. Confusion matrix
Confusion matrix | Real label | Accuracy /% |
---|
| Crime sceneoverview sketch | Self-drawnlocation sketch | Map-screenshotlocation sketch | Negativeclass |
---|
| Crime sceneoverview sketch | 2097 | 39 | 1 | 0 | 98.13 | Predicted label | Self-drawnlocation sketch | 67 | 1177 | 5 | 1 | 94.16 | | Map-screenshotlocation sketch | 3 | 3 | 3012 | 3 | 99.70 | | Negative class | 0 | 0 | 1 | 2715 | 99.96 |
|
查看原文
王凯旋, 李卓容, 王晓宾, 严圣东, 唐云祁. 刑事案件现场图自动分类算法[J]. 激光与光电子学进展, 2020, 57(4): 041009. Kaixuan Wang, Zhuorong Li, Xiaobin Wang, Shengdong Yan, Yunqi Tang. Automated Classification Method for Crime Scene Sketches[J]. Laser & Optoelectronics Progress, 2020, 57(4): 041009.