DESIGN AND IMPLEMENTATION OF JUVENILE FISH COUNTING SYSTEM USING YOLO DETECTION MODEL WITH KALMAN FILTER-BASED TRACKING
In aquaculture industries, fish hatcheries are essential in the production cycle of aquaculture. They are responsible to provide the industry with juvenile fish required for cultivation. In order to achieve high counting accuracy that allows for continuous counting of juvenile fish, a system that ca...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Project |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/87738 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | In aquaculture industries, fish hatcheries are essential in the production cycle of aquaculture. They are responsible to provide the industry with juvenile fish required for cultivation. In order to achieve high counting accuracy that allows for continuous counting of juvenile fish, a system that captures video of fish flowing through a channel is selected. During the development of the counting system, we limit the fish type to catfish with size range of 5-12 cm and tilapia with size range of 7-12 cm. The design of the counting system consists of three subsystems, the object detection module, the object tracking module, and the counting module. The detection module used YOLOv6-N as it performs the best in our use case, 0.9 recall and 0.885 average precision and performs the fastest compared to another YOLO models in our hardware. The tracking module used Improved SORT algorithm, which is a kalman filter-based multi-object tracking algorithm. The counting module used two separate algorithm for catfish and tilapia. When tested on real time counting with different number of fish and three time iterations, the average accuracy of both catfish tilapia, still lacks around 1% from reaching the desired accuracy, only reaching average accuracy of 94.4%. |
|---|