CN202010323629
CN111490823A
微信扫码查看/分享专利
|
摘要:本发明公开一种基于卷积神经网络的可见光成像通信解码方法,该方法在发射端使用RGB‑LED为光源,待发送数据采用OOK编码并分成三个数据流,三个数据流分别驱动RGB‑LED的红色、绿色、蓝色灯珠发出光信号,在接收端采用CMOS传感器摄像头正对RGB‑LED录像来捕捉光信号,然后逐帧提取,从图像中截取条纹分布区域,分别对R通道、G通道和B通道生成平均帧并归一化,接着以有限个条纹为单位裁剪图片,再输入已经训练好的卷积神经网络识别,解码后的数据依序排列组合,还原出原始数据。该方法简单可行,可通过改造现有灯具结合智能手机实现,具有广阔的市场价值。 |
---|---|
Abstract: The invention discloses a visible light imaging communication decoding method based on a convolutional neural network. According to the method, an RGB-LED is used as a light source at a transmitting end; to-be-sent data is encoded by adopting OOK and is divided into three data streams; the three data streams respectively drive red, green and blue lamp beads of the RGB-LED to emit light signals; ata receiving end, a CMOS sensor camera is adopted to directly face an RGB-LED video to capture an optical signal; extracting is performed frame by frame, stripe distribution areas are intercepted fromthe image to respectively generate average frames for the R channel, the G channel and the B channel and normalize the average frames, the image is cut by taking a limited number of stripes as a unit, trained convolutional neural network identification and decoded data are input, the data are sequentially permutated and combined, and original data is restored. The method is simple and feasible, can be realized by combining an existing lamp with a smart phone, and has wide market value. | |
申请人: 深圳市南科信息科技有限公司 | |
Applicant: SOUTH TECH INFORMATION TECH CO LTD | |
地址: 518000 广东省深圳市福田区********(隐藏) | |
发明(设计)人: 刘满喜 王净民 伍文飞 关伟鹏 | |
Inventor: LIU MANXI; WANG JINGMIN; WU WENFEI; GUAN WEIPENG | |
主分类号: H04B10/116(20130101) | |
分类号: H04B10/116(20130101) H04B10/50(20130101) H04B10/516(20130101) H04N5/232(20060101) H04N5/374(20110101) G06K9/62(20060101) G06N3/04(20060101) G06N3/08(20060101) G06T7/90(20170101) | |
2021-03-19 | 授权 |
2020-08-28 | 实质审查的生效IPC(主分类):H04B10/116 |
2020-08-04 | 公开 |
主权项 | 1.一种基于卷积神经网络的可见光成像通信解码方法,其特征在于,包括以下步骤:/n步骤1:在发射端使用RGB-LED为光源,待发送数据采用OOK编码并分成三个数据流,所述三个数据流分别驱动RGB-LED的红色灯珠、绿色灯珠、蓝色灯珠发出光信号;/n步骤2:在接收端采用CMOS传感器摄像头正对RGB-LED录像来捕捉光信号,然后逐帧提取图像,从图像中截取条纹分布区域,分别对R通道、G通道和B通道生成平均帧并归一化,接着以有限个条纹为单位裁剪图片;/n步骤3:取部分标签已知的条纹图片,输入到结构既定的卷积神经网络,以最小化交叉熵损失函数为目标,反复修正卷积神经网络的参数,建立分类器;/n步骤4:将上述剪裁所得的条纹图片输入到已经训练好的卷积神经网络进行识别,将识别结果转化为相应的bit数据,再将数据依序排列组合,依据头部还原发送端发送的数据包。/n | ||||||||||||||||||
公开号 | 111490823A | ||||||||||||||||||
公开日 | 2020-08-04 | ||||||||||||||||||
专利代理机构 | 44384 深圳市中科创为专利代理有限公司 | ||||||||||||||||||
代理人 | 谭雪婷 谢亮 | ||||||||||||||||||
颁证日 | |||||||||||||||||||
优先权 |
|
||||||||||||||||||
国际申请 | |||||||||||||||||||
国际公布 | |||||||||||||||||||
进入国家日期 |
类型 | 阶段 | 文献号 | 公开日期 | 涉及权利要求项 | 相关页数 |
A | SEA | CN106921470A | 20170704 | 1-5 | 全文 |
Y | SEA | CN107490374A | 20171219 | 1-5 | 说明书第[0005]-[0027]段 |
A | SEA | US2018262291A1 | 20180913 | 1-5 | 全文 |
A | SEA | CN110532860A | 20191203 | 1-5 | 全文 |
A | SEA | CN108347281A | 20180731 | 1-4 | 全文 |
类型 | 阶段 | 期刊文摘名称 | 作者 | 标题 | 涉及权利要求项 | 相关页数 |
Y | SEA | 《Signal Processing: Image Communication》 | Heng Zhang 等 | The optical fringe code modulation and recognition algorithm based on visible light communication using convolutional neural network | 1-5 | 第128-140页 |
HENG ZHANG 等: "The optical fringe code modulation and recognition algorithm based on visible light communication using convolutional neural network", 《SIGNAL PROCESSING: IMAGE COMMUNICATION》 | ||||||
类型 | 阶段 | 书名 | 作者 | 标题 | 涉及权利要求项 | 相关页数 |