基于BP神经网络识别汽车前保险杠方法

The Method for Detecting Front Bumper Plastics of Automobile Based on BP Neural Network

交通事故现场中经常发现破碎的汽车前保险杠塑料,如何快速准确地识别保险杠型号,是确定肇事车辆、分析案情发展的重要保障。提出了一种基于BP神经网络的汽车前保险杠塑料识别方法。该方法以汽车前保险杠塑料的近红外光谱数据为基础,通过标准差来分析数据在不同波段的差异程度,据此分为高、中、低差异度3类,使用主成分分析法对3类波段中的数据分别进行降维处理并构成三维数据,使其能够在三维空间内进行区分,提高鲁棒性,并利用该三维数据训练BP神经网络,实现了汽车前保险杠塑料的快速识别,其正确率达到了93.75%,为现场物证的检验提供一种全新、快速的方法。

Broken front bumper plastics are often found in the scene of traffic accidents. How to identify the type of bumper quickly and accurately is an important guarantee to recognize the model of vehicle and analyze the process of the case. A method of plastic identification of front bumper based on BP neural network was proposed. Based on the near-infrared spectroscopy data of the front bumper plastic,the method analyzes the difference of data in different wavebands by standard deviation and divides them into three classes of high,medium and low differences. The data in the class bands are dimensionally reduced to form three-dimensional data by using principal component analysis,enabling them to be distinguished in three-dimensional space,improving robustness, and using this three-dimensional data to train the BP neural network to achieve rapid recognition of front bumper plastics. The correct rate is 93.75%,which provides a new and fast method for the inspection of material evidence.