Microfluidic droplet detection for bio medical application using YOLO with COA based segmentation

Nowadays, biological and biochemical analysis using droplet-based microfluidics has become common practice. This process involves precisely locating the interface between miscible flows and dynamically controlling the sizes of oil-infused droplets in the segmented stream. Miniaturization for point-of-care diagnostics is hampered by the existing optical detection technology because it is costly, time-consuming, and requires a lot of processing to detect droplet contents. So, Optimized deep learning approach is developed to detect microfluidic droplets in bio medical research. To improve image quality, this proposed model pre-processes raw input images with Perona Malik Anisotropic Diffusion and range constrained bi-histogram equalization. Perona Malik Anisotropic Diffusion is used to denoise the input image, and range-limited bi-histogram equalization is used to increase the contrast level of the denoised image. The pre-processed image serves as an input for the droplet segmentation procedure. Image segmentation creates masks or contours for each recognized object in an image, as well as bounding boxes around those objects. To segment the pre-processed image, this current research uses the Optimized You Only Look Once (YOLO v7) technique. Anchor boxes are utilized in YOLO to produce candidate regions as well as anticipate bounding box adjustments and objectness ratings. Cheetah Optimization Algorithm (COA) is used for optimal selection of best anchor boxes in YOLOv7 rather than manually mapping the coordinates of anchor boxes. Proposed optimized deep learning model provides 96% accuracy, 93% precision, and 93.3% recall through a simulation study by comparing the actual and predicted droplets. Thus, the YOLO v7 with COA based segmentation improves the droplet prediction and servers as an efficient tool in bio medical research.