查询结果:   李宁,王雨萱,徐守坤,石林.基于AlexNet的小样本水面漂浮物识别[J].计算机应用与软件,2019,36(2):245 - 251.
中文标题
基于AlexNet的小样本水面漂浮物识别
发表栏目
人工智能与识别
摘要点击数
811
英文标题
RECOGNITION OF FLOATING OBJECTS ON WATER SURFACE WITH SMALL SAMPLE BASED ON ALEXNET
作 者
李宁 王雨萱 徐守坤 石林 Li Ning Wang Yuxuan Xu Shoukun Shi Lin
作者单位
常州大学 信息科学与工程学院 数理学院 江苏 常州 213164 福建省信息处理与智能控制重点实验室 闽江学院   
英文单位
School of Information Science and Engineering, School of Mathematics and Physics, Changzhou University, Changzhou 213164, Jiangsu, China Fujian Key Laboratory of Information Processing and Intelligent Control , Fuzhou 350108, Fujian, China   
关键词
漂浮物图像 小样本 深度学习 AlexNet 光照矫正
Keywords
Image of floating objects Small sample Deep learning AlexNet Light correction
基金项目
闽江学院福建省信息处理与智能控制重点实验室开放课题(MJUKF201740)
作者资料
李宁,副教授,主研领域:数据与信息处理。王雨萱,硕士生。徐守坤,教授。石林,副教授。 。
文章摘要
针对水面漂浮物识别中图像数据量少、噪声影响多,导致识别精度低的问题,采用一种基于深度学习的小样本水面漂浮物识别方法进行水面常见污染物塑料袋与塑料瓶的识别。采用现有大型数据集中的普通塑料袋与塑料瓶图像构建并训练卷积神经网络模型AlexNet;采用梯度下降法对模型进行微调,并用融合的光照矫正法处理待识别图像;将网络识别结果与传统的HOG特征提取方法进行比较。实验结果表明,该方法相较于传统的提取特征方法,对于水面漂浮物的识别率提高近15%。
Abstract
In the recognition of floating objects on water surface, the small amount of image data and the influence of noise leads to the low recognition accuracy. In order to solve this problem, we adopted floating objects on water surface recognition method with small sample based on deep learning to identify common pollutants, plastic bags and plastic bottles. The convolution neural network model AlexNet was constructed and trained from the images of plastic bags and bottles in the existing large data sets. The gradient descent method was used to fine-tune the model, and the fusion light correction method was used to process the image to be recognized. The network recognition results were compared with the traditional HOG feature extraction methods. The experimental results show that compared with the traditional feature extraction method, the method improves the recognition rate of floating objects on the water surface by nearly 15%.
下载PDF全文