AquaWatch

Problem & Description

The AquaWatch project addresses the critical issue of identifying harmful algae and determining toxicity levels in water bodies. Harmful algal blooms pose significant threats to aquatic ecosystems and public health. Motivated by the need for an effective solution, AquaWatch employs advanced image classification techniques to discern various algae species and assess potential environmental risks. The project aims to contribute to water quality monitoring and enhance early warning systems.

Data Source & Data Science Approach

The project draws from diverse datasets, including Harmful Algal Bloom Datasets, Clear Lake Cyanotoxin Issues, Harmful Cyanobacterial Bloom Advisories in Wyoming Waters, and Algal Bloom Sampling Status in Florida Water. Leveraging these datasets, AquaWatch utilizes a two-stepped multi-binary ResNet model for image classification. The model is trained to initially classify algae as non-advisory or advisory and subsequently categorize advisory algae into caution or danger. The approach involves extensive exploratory data analysis (EDA), feature engineering, and the exploration of various model architectures. Feature engineering involves preprocessing and augmenting images, including resizing, pixel normalization, rotation, shifting, flipping, and adjustments to brightness and contrast. 

Evaluation

The performance of AquaWatch is evaluated based on the accuracy and precision of its predictions. Additionally, the project assesses the model's effectiveness in classifying different toxicity levels and its generalization across diverse environmental conditions. The evaluation criteria include metrics such as recall, precision, and F1-score to ensure a comprehensive assessment of the model's capabilities.

Key Learnings & Impact

Throughout the project, key learnings have emerged, emphasizing the importance of robust data preprocessing, diverse dataset representation, and the efficacy of a multi-step ResNet architecture. The impact of AquaWatch lies in its potential to provide valuable insights into harmful algal blooms, enabling timely interventions and mitigating risks to aquatic ecosystems and public health.

Acknowledgements

The success of AquaWatch is indebted to the availability of diverse datasets and the collaborative efforts of contributors to water quality research. Special acknowledgment is extended to the creators of Harmful Algal Bloom Datasets, Clear Lake Cyanotoxin Issues, Harmful Cyanobacterial Bloom Advisories in Wyoming Waters, and Algal Bloom Sampling Status in Florida Water. The project also appreciates the open-source community for the tools and libraries utilized, enhancing the robustness and efficiency of the image classification model.