摘要
为提高在近色背景下果实识别的准确性,减少非结构化因素对识别的影响,提出了基于近红外像机和可见光像机组合捕获多源图像进行融合的方法。首先对已配准的多源图像分别进行非下采样轮廓波变换(NSCT),得到高频系数与低频系数;对高频系数采用压缩融合,并通过CoSaMp恢复融合的高频系数;对低频系数进行小波分解,对分解的高频子带采用绝对值最大法进行融合;低频子带则采用基于几何距离和能量距离加权的融合方法,再通过小波逆变换得到融合的低频系数;最后对融合后的高、低频系数进行NSCT重构得到融合图像。试验结果表明,所设计方法有效地保留了图像的边缘轮廓,突出了图像的细节信息,在客观定量评价指标上均优于其他传统方法,其中与小波变换-非下采样轮廓波变换(DWT-NSCT)方法相比,最大提升达到15.59%。
In order to improve the accuracy of fruit recognition in close background and lower the impact of unstructured factors on recognition, a method of fusion based on multi-source images captured by the combination of near infrared camera and visible light camera is proposed. Firstly, the registered multi-source images are decomposed into high and low frequency coefficients by using NSCT respectively;The decomposition of high frequency coefficients are compressed and fused, and the fused high frequency coefficients are reconstructed by CoSaMp;Then, the low frequency coefficients are decomposed into high and low frequency subbands by DWT, the fusion rule based on absolute maximum method is adopted in high frequency subbands;the low frequency subbands are based on geometrical distance combined with energy distance, and the fused lower frequency coefficients are obtained by inverse transform;Finally, the fusion image is gained by the reconstruction of NSCT;The experimental results show that the proposed method retains the edge contour of the image effectively, highlights the details of the image, and is superior to other traditional methods in objective quantitative evaluation. Compared with DWT-NSCT method, the maximum improvement is 15.59%.
作者
吕继东
吕小俊
王艺洁
徐黎明
马正华
Lv Jidong;Lv Xiaojun;Wang Yijie;Xu Liming;Ma Zhenghua(School of Information and Mathematics,Changzhou University,Changzhou,213164,China;School of Equipment Engineering,Jiangsu Urban and Rural Construction College,Changzhou,213147,China)
出处
《中国农机化学报》
北大核心
2020年第3期141-146,共6页
Journal of Chinese Agricultural Mechanization
基金
江苏省自然科学青年基金项目(BK20140266)
江苏省高等学校自然科学研究项目(17KJB416002)
常州市科技计划资助项目(CJ20179057,CJ20180021)。
