Detection of microalgae objects based on the Improved YOLOv3 model.

Area of Science:

• Marine Biology
• Computational Biology
• Environmental Science

Background:

• Microalgae in ballast water are a primary vector for invasive species introduction.
• Traditional methods for microalgae identification in ballast water face significant limitations.
• Accurate identification is crucial for preventing ecological disruption.

Purpose of the Study:

• To develop an advanced method for identifying microalgae species in ballast water.
• To improve the accuracy and efficiency of microalgae detection using deep learning.
• To address the limitations of existing ballast water monitoring techniques.

Main Methods:

• Utilized an Improved YOLOv3 model with MobileNet as the backbone for feature extraction.
• Incorporated improved spatial pyramid pooling (SPP) to enhance small object detection.
• Optimized the YOLOv3 loss function using the Complete IoU (CIoU) algorithm.
• Experimentally validated the model against established datasets and existing methods.

Main Results:

• The Improved YOLOv3 model achieved an average accuracy of 98.90%.
• Detection efficiency was 8.59% higher compared to the original YOLOv3 model.
• Single image identification time averaged 0.086 seconds, demonstrating high speed.
• The model showed superior performance over existing microalgae identification methods.

Conclusions:

• The Improved YOLOv3 model provides a highly accurate and efficient solution for microalgae identification in ballast water.
• This deep learning approach significantly enhances the ability to detect and manage invasive aquatic species.
• The method offers a promising tool for environmental monitoring and biosecurity efforts.