Research On Automatic Identification Of Dripping Bloodstain Patterns

Bloodstain is one of the most common physical evidence left in the criminal case.Due to the short drying time of blood,it often appears in the form of bloodstain in the crime scene.As an important evidence of the crime scene,bloodstain plays an important role in verifying or overturning the criminal suspect's confession and analyzing the process of the crime.It can also locate the hit point and help to rebuild the crime scene.As a new subject,the current blood trace morphology analysis mainly uses physics and mathematics to carry out quantitative research on the existing blood trace morphology,deduce general rules or mathematical models,and obtain more accurate results,which has become a systematic applied discipline.In order to reduce measurement errors and systematic errors,simplify the procedure of crime scene investigation and broaden the means of case detection and analysis,a convolution neural network(CNN)model is introduced to analyze the dripping bloodstain Patterns.Aiming at the most common dripping bloodstain Patterns in the scene,we use convolution neural network(CNN)model to experimentally analyze the dropping height.The blood source of the sample is anticoagulant whole blood.In the preliminary experiment,the caffenet model was used to classify and train 2400 image samples of simulated dripping of red ink.The train accuracy reached up to 96.7%.The preliminary experiment proved that CNN was feasible in the morphological analysis of dripping bloodstain patterns.On this basis,the experimental study on the morphological recognition of dripping bloodstain Patterns was carried out.The samples of test set were selected from the simulated crime scene.VGG-16 model,ALEXNET model and self-defined convolution neural network model were used to extract and classify 26,000 dripping bloodstain Pattern images with classification heights of 35 cm,50 cm,65 cm and 80 cm,respectively.The train accuracy converges to 99% and that of the optimal model test is 90.94%.The experimental results show that: 1.The fitting accuracy of CNN model is high.The classification model of drip blood trace morphology based on convolution neural network has few measurement errors and system errors,and the recognition rate of dripping height converges to 99%.2.The actual combat simulation prediction accuracy is high,and the actual test accuracy of the dripping height classification model reaches up to 90.94%.It provides reliable reference data for the criminal technical exploration work.3.The research results of the thesis can optimize the analysis process of dripping bloodstain patterns,and upload the original image directly into the CNN model to realize automatic analysis and prediction.The results are fast and effective,and provide technical reference for the police.