| The development of Image Processing technology and Data Mining has brought changes in Safety Precaution and Intelligent Monitoring Industry.Many Digital Image information extraction,processing and understanding technologies have been widely developed and applied.Such as video image object detection,local feature detection,video object tracking and behavior recognition technology.In recent years,more and more researchers begin to explore a way combining image processing and data mining technology to apply in animal behavior understanding research and healthy animal raising,especially on common livestock and poultry such as pig,cattle and chicken.However,research on special animal raising are rare.Also,special raising animals have unique features which different from common livestock.Take porcupine as an example,it has sharp thorn in the back.Although porcupine doesn't offensive in nature,the thorn in the back will stretch when it gets scared which could do harm to the researchers and breeder.And porcupine is nocturnal animal who is active at night time,which brings many obstacles to manual monitoring and lead to omission.If we could find a way to abstract and store porcupine's real time behavior information,combining data mining to explore deep meaning such as porcupine's healthy level or its adaptive level to the raising environment in these information,then we could provide low cost and high accuracy solution to porcupine disease warning and raising environment feedback adjustment.To exp lore a way to solve these problems,we take porcupines in large scale farms as object,focus on porcupine detection and behavior recognition based on digital image processing systems which consist of porcupine detection,continuous frame tracking and porcupine behavior recognition model.At the same time develop prototype system to test the model.With the willing object to assisting porcupine behavior and habits study,to help improve the technical level of artificial breeding of porcupine and provide some thoughts to related research.This article is back by the survey on porcupine scale farm of Guangzhou Shuguang Agricultural Development Limited Company.The research is based on team's decades of accumulation on pig's healthy raising research,through study and reference of recent developments in related research at home and abroad,together with lots of test and algorithm compare.And considering the practical condition of porcupine raising environment we proposed an artificial raised porcupine behavior r ecognition model base on mixture of Gaussian background modelling method,ORB keypoints detection and data mining algorithm.At first,through mixture Gaussian background modeling we get the background and foreground image and marked all the moving contours which containing all the moving porcupines,then we classified the contours with classifiers which get an accuracy of 86.34% of porcupine detection.In order to improve the performance,we introduced local features ORB keypoints as property to training t he data,which improved the detection accuracy up to 93.23%.The result shows our method have a feature of robust and suitable for real time processing.Based on these,we construct a porcupine behavior recognition model which fully considered the distribution of the raising pool and porcupines real activity conditions.Through the model,we successfully achieve the recognition of artificial raised porcupines' 8 basic behaviors which including running,rest,closing,eating,drinking,excretion,biting iron door and biting the water channels total.This model's accuracy to the most frequent behavior rest,running,closing,biting iron door are 91%,88%,86% and 100%.The result shows the model we proposed could back the porcupine intelligent monitoring,artificial raising of porcupines' information collection and porcupine behavior study research. |
PDF Full Text Request