基于深度学习算法的烈性犬检测提醒系统

A Fierce Dog Detection and Reminder System Based on Deep Learning Algorithm

近些年来烈性犬伤人事件屡有发生,通过犬类识别对烈性犬伤人事件进行提前预警成为了当下人工智能的一个研究方向。犬类识别属于深度学习细粒度图像分类的一个典型任务,相比通常的猫狗种类识别等粗颗粒图像识别存在识别率低等技术难题,本文通过人工神经网络扩充训练集、尝试多种卷积神经网络等方式提升了犬类的识别率,并采用其中一种算法成功识别出烈性犬,从而将针对烈性犬伤人事件从“事后追责”变成“提前预警”。本文中的扩充数据集采用普通卷积神经网时有效率为8.299%,采用迁移学习VCG16[1]准确率提升至81.3423%,采用ResNet准确率提升至89.6542%。

In recent years,there are frequent cases of fierce dog injuries.It has become a research direction of AI to give early warning of fierce dog injuries through dog identification which is a typical task of deep learning fine-grained image classification.Compared with the general coarse-grained image recognition such as cat and dog type recognition,there are technical problems such as low recognition rate.In this paper,the recognition rate of dogs is improved by expanding the training set of artificial neural network and trying various convolution neural networks,and one of the algorithms is successfully used to recognize the fierce dogs.Thus,the incident of fierce dog injury will be changed from"after the event"to"early warning".The effective rate of the extended data set in this paper is 8.299%when using ordinary convolutional neural network,the accuracy rate of VCG16 using transfer learning is increased to 81.3423%,and the accuracy rate of ResNet is increased to 89.6542%.