针对当前电力工程现场审计工作缺乏数字化手段支撑,难以准确识别设备、材料类型及数量的问题,提出一种单尺度视网膜(single scale retinex,SSR)算法结合改进YOLOv5s(you only look once version 5 small,YOLOv5s)的电力设备智能检查识...针对当前电力工程现场审计工作缺乏数字化手段支撑,难以准确识别设备、材料类型及数量的问题,提出一种单尺度视网膜(single scale retinex,SSR)算法结合改进YOLOv5s(you only look once version 5 small,YOLOv5s)的电力设备智能检查识别方法。首先,使用SSR算法对数据集中低照度的图片进行增强处理,恢复电力设备色彩与表面细节信息;随后,分析电力设备初始锚框机制,将YOLOv5s初始锚框聚类算法改为K-means++算法,使获得锚框更适用于电力设备识别,提高检测精度;接着,在特征提取阶段,采用动态卷积增加目标检测头,增强模型对目标的识别敏感度,可以更加灵活地处理各种不同的目标;最后,针对电力设备边界框定位不准确的问题,修改YOLOv5s中的损失函数为高效交并比(efficient intersection over union,EIoU)损失函数,进一步提高检测电力设备的精度。通过自主创建多样性电力设备采集图像数据集,将增强后的图像与原图像一并加入数据集,进行试验,试验结果表明,该方法能有效地完成电力设备的自动检测与识别任务,实现对电力工程审计典型要素的快速提取,为电力企业工程数智化审计应用提供了技术支撑。展开更多
针对混合曝光成像算法过程中会出现低曝光处细节丢失且颜色失真饱和度不佳导致视觉观感下降的问题,提出一种多尺度权重评估的MSRCR(Multi-Scale Retinex with Color Restoration,MSRCR)混合曝光融合算法。基于Retinex模型将待融合图像...针对混合曝光成像算法过程中会出现低曝光处细节丢失且颜色失真饱和度不佳导致视觉观感下降的问题,提出一种多尺度权重评估的MSRCR(Multi-Scale Retinex with Color Restoration,MSRCR)混合曝光融合算法。基于Retinex模型将待融合图像分解为亮度分量与反射光分量,对亮度分量结合ACES函数构造光照补偿归一化函数进行处理,对反射光分量加入颜色恢复函数提升色彩细节;分别从曝光量、饱和度、对比度、色域四个尺度设计图像融合权重值,通过多尺度评估优化融合比例;利用Laplacian金字塔融合算法进行多尺度权重融合获得最终图像。实验结果表明,与传统的图像融合算法相比,该算法处理效果较好,有效降低了暗处失真率,提升了视觉信息保真度。展开更多
为实现田间环境下对玉米苗和杂草的高精度实时检测,本文提出一种融合带色彩恢复的多尺度视网膜(Multi-scale retinex with color restoration,MSRCR)增强算法的改进YOLOv4tiny模型。首先,针对田间环境的图像特点采用MSRCR算法进行图像...为实现田间环境下对玉米苗和杂草的高精度实时检测,本文提出一种融合带色彩恢复的多尺度视网膜(Multi-scale retinex with color restoration,MSRCR)增强算法的改进YOLOv4tiny模型。首先,针对田间环境的图像特点采用MSRCR算法进行图像特征增强预处理,提高图像的对比度和细节质量;然后使用Mosaic在线数据增强方式,丰富目标检测背景,提高训练效率和小目标的检测精度;最后对YOLOv4tiny模型使用K-means++聚类算法进行先验框聚类分析和通道剪枝处理。改进和简化后的模型总参数量降低了45.3%,模型占用内存减少了45.8%,平均精度均值(Mean average precision,mAP)提高了2.5个百分点,在Jetson Nano嵌入式平台上平均检测帧耗时减少了22.4%。本文提出的PruneYOLOv4tiny模型与Faster RCNN、YOLOv3tiny、YOLOv43种常用的目标检测模型进行比较,结果表明:PruneYOLOv4tiny的mAP为96.6%,分别比Faster RCNN和YOLOv3tiny高22.1个百分点和3.6个百分点,比YOLOv4低1.2个百分点;模型占用内存为12.2 MB,是Faster RCNN的3.4%,YOLOv3tiny的36.9%,YOLOv4的5%;在Jetson Nano嵌入式平台上平均检测帧耗时为131 ms,分别是YOLOv3tiny和YOLOv4模型的32.1%和7.6%。可知本文提出的优化方法在模型占用内存、检测耗时和检测精度等方面优于其他常用目标检测算法,能够为硬件资源有限的田间精准除草的系统提供可行的实时杂草识别方法。展开更多
According to the characteristics of dynamic firing in pulse coupled neural network (PCNN) and regional configuration in retinal blood vessel network, a new method combined with simplified PCNN and fast 2D-Otsu algorit...According to the characteristics of dynamic firing in pulse coupled neural network (PCNN) and regional configuration in retinal blood vessel network, a new method combined with simplified PCNN and fast 2D-Otsu algorithm was proposed for automated retinal blood vessels segmentation. Firstly, 2D Gaussian matched filter was used to enhance the retinal images and simplified PCNN was employed to segment the blood vessels by firing neighborhood neurons. Then, fast 2D-Otsu algorithm was introduced to search the best segmentation results and iteration times with less computation time. Finally, the whole vessel network was obtained via analyzing the regional connectivity. Experiments implemented on the public Hoover database indicate that this new method gets a 0.803 5 true positive rate and a 0.028 0 false positive rate on an average. According to the test results, compared with Hoover algorithm and method of PCNN and 1D-Otsu, the proposed method shows much better performance.展开更多
DR （diabetic retinopathy） is a most probable reason of blindness in adults, but the only remedy or escape from blindness is that we have to detect DR as early. Several automated screening techniques are used to dete...DR （diabetic retinopathy） is a most probable reason of blindness in adults, but the only remedy or escape from blindness is that we have to detect DR as early. Several automated screening techniques are used to detect individual lesions in the retina. Still it takes more dependency of time and experts. To overcome those problems and also automatically detect DR in easier and faster way, we took into soft computing approaches in our proposed work. Our proposed work will discuss several amounts of soft computing algorithms, it can detect DR features （landmark and retinal lesions） in an easy manner. Processes includes are： （1） Pre-processing; （2） Optic disc localization and segmentation; （3） Localization of fovea; （4） Blood vessel segmentation; （5） Feature extraction; （6） Feature selection; Finally （7） detection of diabetic retinopathy stages （mild, moderate, severe and PDR）. Our experimental results based on Matlab simulation and it takes databases of STARE and DRIVE. Proposed effective soft computing approaches should improve the sensitivity, specificity and accuracy.展开更多
文摘针对当前电力工程现场审计工作缺乏数字化手段支撑,难以准确识别设备、材料类型及数量的问题,提出一种单尺度视网膜(single scale retinex,SSR)算法结合改进YOLOv5s(you only look once version 5 small,YOLOv5s)的电力设备智能检查识别方法。首先,使用SSR算法对数据集中低照度的图片进行增强处理,恢复电力设备色彩与表面细节信息;随后,分析电力设备初始锚框机制,将YOLOv5s初始锚框聚类算法改为K-means++算法,使获得锚框更适用于电力设备识别,提高检测精度;接着,在特征提取阶段,采用动态卷积增加目标检测头,增强模型对目标的识别敏感度,可以更加灵活地处理各种不同的目标;最后,针对电力设备边界框定位不准确的问题,修改YOLOv5s中的损失函数为高效交并比(efficient intersection over union,EIoU)损失函数,进一步提高检测电力设备的精度。通过自主创建多样性电力设备采集图像数据集,将增强后的图像与原图像一并加入数据集,进行试验,试验结果表明,该方法能有效地完成电力设备的自动检测与识别任务,实现对电力工程审计典型要素的快速提取,为电力企业工程数智化审计应用提供了技术支撑。
文摘针对混合曝光成像算法过程中会出现低曝光处细节丢失且颜色失真饱和度不佳导致视觉观感下降的问题,提出一种多尺度权重评估的MSRCR(Multi-Scale Retinex with Color Restoration,MSRCR)混合曝光融合算法。基于Retinex模型将待融合图像分解为亮度分量与反射光分量,对亮度分量结合ACES函数构造光照补偿归一化函数进行处理,对反射光分量加入颜色恢复函数提升色彩细节;分别从曝光量、饱和度、对比度、色域四个尺度设计图像融合权重值,通过多尺度评估优化融合比例;利用Laplacian金字塔融合算法进行多尺度权重融合获得最终图像。实验结果表明,与传统的图像融合算法相比,该算法处理效果较好,有效降低了暗处失真率,提升了视觉信息保真度。
文摘为实现田间环境下对玉米苗和杂草的高精度实时检测,本文提出一种融合带色彩恢复的多尺度视网膜(Multi-scale retinex with color restoration,MSRCR)增强算法的改进YOLOv4tiny模型。首先,针对田间环境的图像特点采用MSRCR算法进行图像特征增强预处理,提高图像的对比度和细节质量;然后使用Mosaic在线数据增强方式,丰富目标检测背景,提高训练效率和小目标的检测精度;最后对YOLOv4tiny模型使用K-means++聚类算法进行先验框聚类分析和通道剪枝处理。改进和简化后的模型总参数量降低了45.3%,模型占用内存减少了45.8%,平均精度均值(Mean average precision,mAP)提高了2.5个百分点,在Jetson Nano嵌入式平台上平均检测帧耗时减少了22.4%。本文提出的PruneYOLOv4tiny模型与Faster RCNN、YOLOv3tiny、YOLOv43种常用的目标检测模型进行比较,结果表明:PruneYOLOv4tiny的mAP为96.6%,分别比Faster RCNN和YOLOv3tiny高22.1个百分点和3.6个百分点,比YOLOv4低1.2个百分点;模型占用内存为12.2 MB,是Faster RCNN的3.4%,YOLOv3tiny的36.9%,YOLOv4的5%;在Jetson Nano嵌入式平台上平均检测帧耗时为131 ms,分别是YOLOv3tiny和YOLOv4模型的32.1%和7.6%。可知本文提出的优化方法在模型占用内存、检测耗时和检测精度等方面优于其他常用目标检测算法,能够为硬件资源有限的田间精准除草的系统提供可行的实时杂草识别方法。
基金Project (60872081) supported by the National Natural Science Foundation of ChinaProject (50051) supported by the Program for New Century Excellent Talents in UniversityProject (4092034) supported by the Natural Science Foundation of Beijing
文摘According to the characteristics of dynamic firing in pulse coupled neural network (PCNN) and regional configuration in retinal blood vessel network, a new method combined with simplified PCNN and fast 2D-Otsu algorithm was proposed for automated retinal blood vessels segmentation. Firstly, 2D Gaussian matched filter was used to enhance the retinal images and simplified PCNN was employed to segment the blood vessels by firing neighborhood neurons. Then, fast 2D-Otsu algorithm was introduced to search the best segmentation results and iteration times with less computation time. Finally, the whole vessel network was obtained via analyzing the regional connectivity. Experiments implemented on the public Hoover database indicate that this new method gets a 0.803 5 true positive rate and a 0.028 0 false positive rate on an average. According to the test results, compared with Hoover algorithm and method of PCNN and 1D-Otsu, the proposed method shows much better performance.
文摘DR （diabetic retinopathy） is a most probable reason of blindness in adults, but the only remedy or escape from blindness is that we have to detect DR as early. Several automated screening techniques are used to detect individual lesions in the retina. Still it takes more dependency of time and experts. To overcome those problems and also automatically detect DR in easier and faster way, we took into soft computing approaches in our proposed work. Our proposed work will discuss several amounts of soft computing algorithms, it can detect DR features （landmark and retinal lesions） in an easy manner. Processes includes are： （1） Pre-processing; （2） Optic disc localization and segmentation; （3） Localization of fovea; （4） Blood vessel segmentation; （5） Feature extraction; （6） Feature selection; Finally （7） detection of diabetic retinopathy stages （mild, moderate, severe and PDR）. Our experimental results based on Matlab simulation and it takes databases of STARE and DRIVE. Proposed effective soft computing approaches should improve the sensitivity, specificity and accuracy.
