Research On Snake Image Classification Based On Deep Learning

Snakes are a kind of common reptiles,which can be divided into poisonous snakes and non-poisonous snakes.It happens from time to time that people get bitten by poisonous snakes.After being bitten by a poisonous snake,the patient needs to be treated by a kind of antivenom corresponding to this kind of poisonous snakes.In our country,many species of snakes are protected animals.Some species of snakes have some economic value in some fields such as medicine and pharmacy.In the recent years,the technology of deep learning developed rapidly.Some deep learning techniques such as convolutional neural networks and capsule neural networks have been applied widely in some fields such as computer vision and natural language processing.At present,there only exists several research work about sna:ke image classification.The differences between different species of snakes in the aspects such as color,marking,shape or posture make it possible to classify snake images.It's a natural idea to utilize deep learning techniques to solve the problem of snake image classification.However,one difficult point of this problem is that there exist few public snake image datasets with enough sizes,especially image datasets of common snakes in China.Based on this background,this thesis sets up a snake image dataset named CHINESESNAKES,which consists of 10336 images of ten species of common snakes in China,including Bungarus multicinctus,Trimeresurus stejnegeri,Naja atra,Deinagkistrodon,Elaphe carinata,Python bivittatus,Elaphe mandarinus,Boa constrictor,Rhabdophis subminiatus and Ramphotyphlops braminus.This thesis analyzes the feasibility and difficult points of snake image classification as well.Based on the CHINESESNAKES snake image dataset,this thesis utilizes convolutional neural networks techniques,which is a kind of deep learning techniques,to study the problem of snake image classification.Aiming at solving the problem of snake image classification,this thesis designs a new convolutional neural network architecture named BRC.whose classification accuracy on the CHINESESNAKES snake image dataset achieves 89.061 percent.Finally,this thesis designs and realizes a system for snake image recognition based on Android operating system.