Underwater Fish Identification in Real-Time using Convolutional Neural Network

• Main Authors: Aruna, S.K., Deepa, N., Devi, T
• Format: Conference Proceeding
• Language: English
• Subjects: Artificial Intelligence; Deep learning; Fish; Machine Learning; Manuals; Production; Real-time systems; Soft sensors; Statistical analysis; Tracking Deep Learning; Underwater object detection
• ISSN: 2768-5330
• DOI: 10.1109/ICICCS56967.2023.10142531

Artificial Intelligence (AD is the wide application that learns the problem and features by given data and processes the data like the human brain. When a computer program imitates a characteristic of the human brain that is considered "innovator." Among the methods are statistical methods, methods for artificial intelligence, and traditional order to verify the validity. The expansion of AI is also related to virtually infinite storage and an abundance of data, including exchanges, geospatial information, video files, photos, text messages, and audio files. Machine learning is divided into deep learning and deep learning is primarily divided into numerous layers of neural networks. This pattern gives it the ability to learn a lot of information and attempt to replicate the brain function. Increasing the efficiency by attaching more covert layers can be beneficial. It is used to gather the data and transfer the data. Aquaculture production has grown into a barrier to the growth of fish culture and the counting operation represents one of the problems experienced during the spawning process. Previous studies have primarily relied on the application of manual and automated counting techniques, which has prevented it from producing accurate results. The proposed method offers a promising method for enhancing image detection by combining the IoT techniques. The image data were divided into three categories: low frequency, intermediate density, as well as high frequency. The proposed method has used 8200 images to train and 2500 images for verification. Only the data relevant data sources were used during the train and verification phase in order to find the proper parameters and create a better VGG19 parameter calibration strategy. Consequently, the improved VGG19 model can achieve an accuracy of 98%.