Prediction Of Cell Density Of Several Bait Algae Based On Mobile Phone Images

Microalgae is an important bait for larvae such as shellfish and crustaceans in the process of large-scale agricultural and aquaculture seed production.The success or failure of its intensive cultivation can almost determine the success or failure of seedling raising.In addition,microalgae are rich in protein,amino acids,highly unsaturated fatty acids,pigments,and a variety of biologically active substances.They have become an important source of food,medicine,feed and fuel.Therefore,many research institutions have carried out large-scale intensive cultivation.In these intensive culture processes,counting the density of microalgae cells is the most common but time-consuming task.Therefore,convenient measurement methods can solve the cumbersome measurement problems faced in microalgae research or production practice.In this study,Chlorella vulgaris,Nannochloris oculata,Tetrasemis helgolandica,and Nitzschia closterium were used as the research objects.The 4 species of algae were taken with mobile phones and the RGB color values of the pictures were digitally extracted.,And converted to the other three color spaces,analyzed and established a prediction model of different algae cell densities,to provide a reference for the rapid monitoring of intensive algae cultivation in the future.This paper analyzed the color component characteristics of 4 species of algae at different cell densities and in 4 color spaces.The results showed that the algae cell density had a significant impact on the color component changes of different algae.In the RGB color space of Platymona,the color components R,G,and B values all had a significant impact on the concentration changes.After linear fitting,when the OD<0.83,the correlation coefficients between the R and G values on the concentration changes were the highest,The equations were R =-0.8736×OD + 0.726(R2 = 0.889),G =-0.7388×OD + 0.7545(R2 = 0.8651);when OD<0.22,the correlation coefficient of the B value reached the highest,and the regression equation B =-3.5056 × OD + 0.6924(R 2 = 0.8035).In the HLS color space of Microchlorococcus,the color components H,L,and S values had significant effects on the concentration changes.After linear fitting,when 0.16<OD<1,the regression equation H =28.967×OD + 55.796 was selected.(R2 = 0.84);when OD<0.28,chose the regression equation L = 2.3797×OD + 0.0335(R2 = 0.9662);when OD<0.45,used the regression equation S =-1.3701×OD + 0.7115(R2 = 0.9443).In the RGB color space of Nitzschia lunata,the color components G and B havd a significant effect on the concentration change.When OD<0.45 after linear fitting,used the regression equation G =-0.8567×OD + 0.7153(R2 = 0.9398);When OD<0.28,chose the regression equation B =-2.5725×OD + 0.6969(R2 =0.9291).In the RGB color space of Chlorella,the R value,G value and B value of the color components had significant effects on the concentration change.After linear fitting,when OD<0.94,the regression equation was selected as R =-0.6646×OD + 0.7158(R2 = 0.9015);when 0.03<OD<1,used the regression equation G =-0.4872×OD + 0.7586(R2 = 0.8574);when OD<0.25,chose the regression equation B =-2.9256×OD + 0.7081(R2 = 0.9659).In this paper,6 different light levels are studied,and the influence of different light changes on each component under 4 color spaces is discussed.According to the analyzed data,the spss software was used to analyze the correlation between the concentration of algae and the RGB and HLS color space of the picture.On this basis,a full subset regression model of the algae concentration and the component values of the RGB and HLS color space was established.The interactive and non-interactive effects of color space variables on the determination of substance concentration are analyzed,and a complete subset regression model is established;finally,on the basis of residual analysis,the concept of relative residual is introduced to analyze the error of the obtained model.The results show that the full subset regression model has a higher correlation coefficient and a smaller residual variance,and the resulting model is very concise.