Research And Implementation Of Algorithms For Evaluating Animal Pain Based On Images

Spontaneous pain symptoms in animals are related to chronic pain symptoms in humans,which can better reflect human perception of pain.When developing new analgesics in the field of biomedicine,rodents such as mice are often used to sense and assess pain.Considering that rodent and other mammalian facial salamander characteristics are closely related to their pain levels,researchers usually measure the degree of pain by observing rodent facial salamander characteristics.In this thesis,the facial pain characteristics in mouse images are used as research objects,and deep learning methods are used to realize automatic real-time assessment of pain levels.The specific research work is as follows:1.According to the Mouse Grimace Scale(MGS),a mouse image dataset was produced,which can be used for training and testing of target detection models.The facial pain characteristics of the mice(narrow orbital area,nose bulge,ears pulled back,whiskers backwards)were synthesized as labels for the training data set.2.In order to solve the problem of artificially scoring lots of animal images,which is labor-intensive and subjective,this thesis proposes a method for automatically judging rat pain based on a Single Shot Multi Box Detector(SSD).This method can accurately determine whether the mouse is in pain or innocuous state,and it is applied to 3 different data sets.The experimental results show that the accuracy rate is as high as 95% under the same data set with a confidence level of 0.9,which indicate the SSD model can effectively evaluate the pain state of mice.3.In order to more accurately measure the degree of pain in mice,this thesis proposes an improved first-order rapid detection model(You Only Look Once,YOLOv3).The introduction of label smoothing parameters improves the robustness and generalization of the model;By introducing the Focal loss function,the prediction is more accurate;Merging batch normalization layers(BN)into convolutional layers improves the rate of network convergence.The experimental results show that the automatic measurement method of animal pain based on the improved YOLOv3 model can well applied to the mouse dataset.The accuracy rate is 84% with an absolute error of 1.5.It can evaluate 62 mouse pictures per second with high accuracy.It has high accuracy and real-time performance.Not only can the mouse's face be accurately located,but also different color mouse species can accurately measure their pain.