加权高斯减法拟合的亮度可控土壤图像增强  

Controllable Brightness Enhancement of the Soil Image Based on Weighted Gaussian Subtraction Fitting

作　　者：曾绍华[1,2] 赵秉渝 王帅 陈亚楠 朱德利 ZENG Shaohua;ZHAO Bingyu;WANG Shuai;CHEN Yanan;ZHU Deli(College of Computer and Information Science,Chongqing Normal University,Chongqing 401331,China;Chongqing Research Center on Engineer Technology of Digital Agricultural&Services,Chongqing 401331,China;Chongqing Master Station of Agricultural Technology Promotion,Chongqing 400014,China;Chongqing Wanzhou District Station of Soil Fertilizer and Agricultural Ecological Protection,Chongqing 404199,China)

机构地区：[1]重庆师范大学计算机与信息科学学院,重庆401331 [2]重庆市数字农业服务工程技术研究中心,重庆401331 [3]重庆市农业技术推广总站,重庆400014 [4]重庆市万州区土肥与农业生态保护站,重庆404199

出　　处：《光子学报》2022年第4期348-367,共20页Acta Photonica Sinica

基　　金：重庆市高校创新研究群体项目(No.CXQT20015);重庆市教委科学技术研究重点项目(No.KJZD-201900505);重庆师范大学研究生科研创新项目(No.YKC20032)。

摘　　要：为了将野外不同光照环境机器视觉采集的土壤图迁移到特定亮度,消除成像条件不一致对紫色土土种识别的影响,提出土壤图像亮度可控增强算法。其方法是构建左局部区域的高斯拟合优化模型对土壤图像Y分量的直方图进行拟合,求拟合剩余部分,继续对剩余部分的左局部区域高斯拟合,直到剩余部分足够小,获得多次高斯拟合的加权拟合曲线;然后在加权高斯减法拟合曲线中引入目标亮度实现图像亮度迁移,完成土壤图像亮度可控增强;最后根据色比不变性原理,对U、V分量颜色校正,获得土壤增强图像。实验结果显示:以原图亮度为基点,正负30个亮度灰度级为所提算法的有效增强变换范围;所提算法比目前文献可查到的仅有的2个可控亮度图像增强对比算法的土壤图像增强变换的精度高,失真度小。With the application of artificial intelligence in agriculture,the requirement of applying machine vision in the field to identify soil species has been raised. Different natural light will bring different images when soil images are collected by machine vision in the field,and it will affect soil species recognition. For refraining from this influence,one method is to collect completely images of soils under a variety of different natural lighting conditions. However,the acquisition of soil images in natural environments can be limited by natural conditions,time and economic costs,and it is difficult to implement. Thus,it may be an effective method that the soil image is converted to be similar to those real soil images that collected in the specific lighting environments,and it can eliminate the influence of inconsistent sunshine environments to improve the accuracy of soil species recognition. The main work of this paper is as follows. Multiple Gaussian fitting of brightness histogram of soil image is realized. Through studying and analyzing soil image,it is found that its Y component histogram is a skewness distribution and its left parts is similar to the left local area of a Gaussian curve,and the remainder that the Y component histogram is fitted by Gaussian curve still remains the features that its left parts is closed to the left local area of a Gaussian curve until the remainder becomes white noise. So the Y component histogram of a soil image can be fitted by several Gaussian curves. Based on the above ideas,an optimization model is established to fit the Gaussian curve of its left local area. Then,the fitting residual is computed and the next Gaussian fitting is executed until the fitting residual is small enough. The weighted fitting curve of multiple Gaussian fitting and weighted Gaussian subtractive fitting algorithm are obtained. Controllable brightness enhancement algorithm of soil image based on weighted Gaussian subtractive fitting is proposed. The target brightness is introduced into the weigh

关 键 词：图像增强 亮度可控 加权高斯减法拟合 土壤图像 机器视觉 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]