Research On The Measurement Method Of Microbubble Size And Quantity

Oxygenation is a key aspect of aquaculture,and microbubble aeration technology has the advantage of high surface area to volume ratio and high density compared to other aeration technologies.As a result,it has higher efficiency and great practical value.To maintain dissolved oxygen levels in water over the long term,optimal bubble size,density,and other characteristics are necessary.This requires optimizing aeration equipment parameters such as ventilation pressure,flow rate,and micropore size.However,due to the lack of accurate methods for detecting microbubble characteristics,there is a lack of a solid theoretical basis for optimizing aeration parameters and designing aeration equipment.This paper focuses on microbubbles generated by electrolyzing water and uses neural networks and semantic segmentation models to extract features such as size and quantity of underwater microbubbles,providing guidance for optimizing the design of microbubble aeration equipment.The main research work of this paper is as follows:(1)In response to the detection problem of microbubble size and quantity,microbubble images were captured using a microscope camera,and FCN and U-Net networks were used for bubble segmentation to compare segmentation effects.The results showed that U-Net had higher segmentation accuracy,with average intersection-over-union,average pixel accuracy,Dice similarity coefficient,and pixel accuracy of 93.29%,96.49%,96.42%,and 99.69%,respectively,satisfying the requirements for fitting bubble size.Then,the Hough transform was used to fit the bubble,obtaining pixel size information,which was then calibrated using a glass microscope ruler to determine the ratio between pixel size and actual size,thus obtaining a measurement of bubble diameter.(2)To solve the problem of counting high-density bubbles,Mask-Rcnn was used to count bubbles that overlap or stick together severely.The crowd density detection algorithm based on regression was also transferred to bubble quantity detection.The counting accuracy of bubbles by CAN,CSRNet,Mask-Rcnn,and U-Net-Hough was compared.CAN had a high accuracy rate of bubble counting based on density map regression,with MAE and RMSE of 3.69 and 5.36,respectively.This method not only accurately and effectively counts high-density bubbles,but also has lower requirements for the quality of bubble images,enabling detection of high-density bubble images taken with low-magnification lenses.(3)The influence of electrolysis current on bubble quantity was explored.Based on the electrolysis method,a microbubble generation device was built,and the size probability density distribution function of bubbles generated by different molybdenum wires was obtained through semantic segmentation.Combined with Faraday’s law and Stokes’ law,a calculation formula for the relationship between bubble quantity and current was derived,which showed that bubble quantity was positively correlated with current.(4)A model for classifying the size and quantity of bubbles based on backscatter echoes has been constructed.Based on Rayleigh and Mie scattering theories,the scattering effect of bubbles on light was analyzed,and the light scattering process of a single bubble was simulated to obtain the light scattering intensity distribution of bubbles,theoretically demonstrating the correlation between backscatter echoes and bubble characteristics.A classification method based on CNN was proposed,using one-dimensional convolution kernels to classify waveform signals,and collecting laser echo samples of 10 different bubble sizes and densities in gas-liquid mixtures as the dataset for training the classification model.To compare the classification performance of the model,15 time-domain features of the echoes were extracted,and the top 4 principal components with a cumulative contribution rate of 98.47% were selected using the KPCA algorithm for training set construction,and the traditional classification method was trained based on this dataset.The results showed that the CNN-based classification method had a higher average accuracy of 99.41%,which was 5.11% higher than the highest classification accuracy of SVM in traditional machine learning.Since laser echo collection is easier to implement than microscopic image collection,and the CNN-based classification method does not require manual feature selection,this method can provide an accurate and convenient means for microbubble feature detection.