Low-Cost Microalgae Cell Concentration Estimation in Hydrochemistry Applications Using Computer Vision

Abstract

Accurate and efficient estimation of microalgae cell concentration is critical for applications in hydrochemical monitoring, biofuel production, pharmaceuticals, and ecological studies. Traditional methods, such as manual counting with a hemocytometer, are time-consuming and prone to human error, while automated systems are often costly and require extensive training data. This paper presents a low-cost, automated approach for estimating cell concentration in Chlorella vulgaris suspensions using classical computer vision techniques. The proposed method eliminates the need for deep learning by leveraging the Hough circle transform to detect and count cells in microscope images, combined with a conversion factor to translate pixel measurements into metric units for direct concentration calculation (cells/mL). Validation against manual hemocytometer counts demonstrated strong agreement, with a Pearson correlation coefficient of 0.96 and a mean percentage difference of 17.96%. The system achieves rapid processing (under 30 s per image) and offers interpretability, allowing specialists to verify results visually. Key advantages include affordability, minimal hardware requirements, and adaptability to other microbiological applications. Limitations, such as sensitivity to cell clumping and impurities, are discussed. This work provides a practical, accessible solution for laboratories lacking expensive automated equipment, bridging the gap between manual methods and high-end technologies.

Keywords: cell concentration; cell segmentation; computer vision; microalgae; microscopy.

Figure 1

Scheme of differences between classical manual cell concentration estimation approach and proposed automatic approach.

Figure 2

Scheme of proposed approach for cell concentration calculation.

Figure 3

Example of auto-build report with image volume calculation for first equipment set (high-quality image—4032 × 3024).

Figure 4

Instrumental setup for wet laboratory experiments for data collection: (a) digital microscope; (b) cultivation chamber with orbital shakers, LED illumination, and temperature regulation; (c) Goryaev chambers.

Figure 5

(a) MAE metric dependence from cells number on image—there is no significant correlation; (b) IoU metric dependence from cells number on image.

Figure 6

Example of StarDist pre-trained model prediction and proposed CV algorithm prediction of cells mask (black color). Color scheme changes inside model.

Figure 7

Comparison of segmentation outputs from YOLO and U-Net models trained on limited data versus ground-truth annotations. The degraded mask quality demonstrates the data requirements for deep learning approaches.

Figure 8

Validation of automatic cell concentration calculation system with expert values. Comparison presented with diluted suspension for clarity when comparing absolute values of different samples.

Figure 9

Examples of automatically generated cell markings compared with manual annotations for Samples 10 and 14, representing low and high cell concentrations, respectively.

Figure 10

Variation in cell concentration between expert calculations (per 9 chamber squares) and the automated system (per 10 images).

Figure 11

Comparison of expert and automated cell counting methods for (A) low-magnification (n = 15) and (B) high-magnification (n = 21) setups. Error bars in (A) represent the distribution of values among squares/images. Dashed lines indicate perfect agreement.

Figure 12

Comparison of processing times between automated and manual cell counting methods. Time measurements (in seconds) are shown for both expert and automated approaches.

Figure 13

System performance on degraded images: foreign object interference, uneven illumination, and focal blurring.

Figure 14

Analysis of percentage error dependency on agglomeration rate of suspension samples.

Figure 15

Examples of fragments of processed images with clots and their annotations.

Figure 16

Representative examples demonstrating system limitations: (1) cell aggregates in poorly mixed suspensions (top row), (2) substrate contamination from alternative growth media (middle row), and (3) concentration artifacts from over-dilution (upper) and under-dilution (lower) samples.

Figure 17

High-magnification images contributing to elevated error rates: Samples 11 and 5 exhibit cell clumping, while Sample 6 contains substrate impurities.