Research On The Detection Of Drip Liquid Level Based On Vision

In the medical field,sick infusion is a common medical phenomenon.During the infusion process,it is necessary to always pay attention to whether the liquid in the drip bottle has been infused,so as to avoid the phenomenon of air entering the blood vessel or blood backflow due to the failure to deal with it in time,Triggering a medical accident.The infusion process takes a long time and consumes effort.In response to this problem,the current research is mostly based on the sensor device to detect the liquid level in the drip bottle,that is,directly or indirectly add an auxiliary sensing device around the container,many containers can not be reused,the cost is too high,there are limitations.In this paper,based on machine vision method to achieve the detection of liquid level line area in drip bottle.The main contents of this study include:1.Segment the foreground image to narrow the detection range.In the research of drip level detection,the collected image is used in the trained convolutional neural network model.The collected drip bottle image has a complex background.In this paper,the Grab Cut algorithm is used to obtain the foreground image of the drip bottle through interaction.After interactive segmentation The foreground image features of the drip bottle are limited to the inside of the bottle,which reduces the detection range,which is convenient for the final detection of the level line area in the drip bottle based on the convolutional neural network model.2.Improve the guide filtering algorithm for denoising.For the research of drip liquid level detection,the foreground image of the drip bottle obtained by the Grab Cut algorithm has the problems of edge depression and spur,combined with the improved guide filter algorithm to denoise the image,the improved guide filter algorithm mainly divides the image to obtain The foreground target image is binarized as the mask image of the guide filter,combined with the original image for denoising,and the experimental results are compared and analyzed with the same type of filter.3.Liquid level detection of drip bottle based on vision.In the research of liquid level detection,most of the traditional methods use sensing devices.In this paper,the vision-based deep learning algorithm is applied to the liquid level detection of drip bottles.That is,the convolutional neural network model after parameter optimization is applied to a large number of drip bottle image data Set to perform training and learning,the Grab Cutsegmentation and then guided filtering processing of the drip bottle image through the trained convolutional neural network model to detect the level line feature area.Experiments have shown that the problem of edge depression and spurs after Grab Cut segmentation,combined with the improved guided filtering method of Grab Cut,is superior to other comparison methods in peak signal-to-noise ratio,structural similarity,and average running time,and has obvious improvements;denoising The refined image of the drip bottle of the front sight is iterated through the trained Faster-RCNN network model for different times,comparing the true positive recognition rate,AP,average running time and the target confidence in the detection effect map.It is found that the learning rate is 0.001,iterative When the number of times is 10000,the running time is average,the true positive recognition rate is the highest,the AP value is the largest,and the target confidence in the detection effect diagram is the highest.At this time,the effect is the best.