Booth Id:
ROBO090
Category:
Robotics and Intelligent Machines
Year:
2021
Finalist Names:
Soni, Rishi (School: Cleveland High School)
Abstract:
The classification of fish through visual inspection can often be a time-consuming and fallible process, and thus, there is a necessity for a fish classification method with greater accuracy and efficiency. This study aimed to develop a fish image classification system for 16 species of fish found in Mouse Creek, Cleveland, TN using machine learning. An initial dataset of 189 high-quality images was expanded using data augmentation and modified with color preprocessing, producing a diverse dataset of 1512 images. Multiple machine learning algorithms were tested using random 80/20 splits of the dataset after Wolfram language built-in feature extraction was performed. Ultimately, this study achieved a machine learning model based on SVM classification that can classify 16 species of fish nearly instantaneously with 99.01% accuracy. This model strongly outperforms alternative image classification tools, making it one of the only effective freshwater fish classification systems for urban stream settings. The system was eventually deployed as a public web app in order to make it universally accessible.