IMPLEMENTATION OF CPU AND NPU BASED COMPUTING HARDWARE FOR REALTIME INFERENCE AND DESIGN AND IMPLEMENTATION OF GRAPHICAL USER INTERFACE FOR AUTOMATIC JUVENILE FISH COUNTER

• Main Author: Fadhil Yanuarsyah, Raihan
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/87884
• Institution: Institut Teknologi Bandung

Manual calculation of juvenile fish requires a lot of time and energy and has inaccuracies that are detrimental to fish farmers. A computer vision-based automatic juvenile fish counter is designed to overcome these problems. This device is designed with a calculation speed specification of 30,000 fish per hour with an accuracy of 95-97%. The computing hardware of this device is implemented using a Single Board Computer Orange Pi 5 Pro which has a Neural Processing Unit (NPU) to accelerate the inference of the object detection model. The use of two multicore NPUs resulted in an average processing time of 0.015 seconds per frame which met the target requirement of 0.0198 seconds per frame. The device is equipped with a graphical user interface using the wxPython framework which was chosen because it has the lowest CPU usage compared to other alternatives. The test results show that the counting accuracy varies based on the number of samples, with an accuracy of 92-94% for 10-50 fish and 95-96% for 100-500 fish. The decrease in accuracy for small sample sizes is due to juvenile fish getting stuck which causes errors in the tracking algorithm. Nonetheless, the system successfully met the targeted processing speed specifications and provided a graphical user interface.