Research On The Type Analysis Of Pliers Cut Marks Based On Detail Image Classification

The pliers cut marks are one of the common physical evidences in criminal cases.The characteristics of this kind of marks can reflect the types of crime tools and the professional characteristics of the suspects.The results of type analysis are often used as clues and basis for finding crime tools and locking offenders in crime scene investigations.Investigators mainly use morphological and statistical methods to perform qualitative and quantitative analysis of pliers cut marks.They summarize the characteristics and use them to guide the source tracing work.Aiming at the problems of low level of intelligent identification of trace evidence inspection and strong subjectivity of manual inspection,this article introduces convolutional neural network to classify detail images and realize the type analysis of pliers cut marks.In this paper,the morphological characteristics of 10 common tools are recorded by using a detail photography method,such as bolt cutters,cable pliers and locking pliers.Experimental research is performed by using convolutional neural networks.Specific experimental method:In the pre-experiment,Lenet is used to train and model detailed images of 4 types of tools.The model iterates 60,000 times with an accuracy rate of 97.78%,which shows that the convolutional neural network is feasible for the type analysis of pliers cut marks.On this basis,the type analysis of ten types of tools begins.Four detailed image sets of different quality are collected and processed.For each set,there are 80,000 training images and 20,000 validation images,a total of 100,000.Four sets totaling 400,000.The test set is 10,000 each,and four are a total of 40,000 images.The Lenet,AlexNet and GoogLenet were used for feature extraction and classification of the image set.12 models with different performance were obtained.The output parameters of the 12 models were compared and analyzed to select the best model.Experiments have improved the best model,deepened the network layers,and optimized the model parameters.The accuracy of the improved model is 95.1%.Using Python to visualize the process of improved model feature extraction,which confirms the reliability of feature extraction.At the same time,a traditional test identification table is made for testing experiments.To ensure that the model can provide a reliable technical reference,confusion matrix is used to compare and analyze the advantages and disadvantages between traditional manual testing and model identification.The experimental results show that: 1.The improved convolutional neural network model has high fitting accuracy and an average accuracy rate of 90%.Using this model to classify detail images,it can complete the type analysis of the pliers cut marks.2.Model identification and traditional manual identification have their own advantages.The model has a fast identification speed and a uniform distribution of false positives,which is suitable for rapidscreening at the initial stage.Traditional manual identification is slower and the false positives are concentrated.It is more suitable for subsequent inspections.3.The model can directly classify the detail images of pliers cut marks.It can realize automatic identification,and provide intelligent evidence analysis methods for marks analysis.This article breaks the traditional visual identification method of pliers cut marks,and innovatively uses deep learning to realize type analysis of pliers cut marks.The paper builds a convolutional neural network model,which has achieved good classification results on self-built image datasets.It can provide a new automatic classification and identification method,which has important practical significance in the actual combat of public security,such as identifying facts and inferring crime tools.