Yolov4-tiny and Spatial Pyramid Pooling for Detecting Head and Tail of Fish

Prasetyo, Eko and Suciati, Nanik and Fatichah, Chastine (0029) Yolov4-tiny and Spatial Pyramid Pooling for Detecting Head and Tail of Fish. International Conference on Artificial Intelligence and Computer Science Technology (ICAICST). ISSN 978-1-6654-2402-8

	Text icaicst (cek plagiasi) (1).pdf Download (1MB)
	Text Conference 2 - Yolov4-tiny+SPP (1).pdf Download (598kB)
	Text 2021-ICAICST-(2021)-Yolov4-tiny and Spatial Pyramid Pooling for Detecting Head and Tail of Fish (1).pdf Download (337kB)

Abstract

Touchless fish freshness monitoring is an appropriate approach to avoid the destruction of the fish due to fingering or bacterial contamination from the hands of consumers. A freshness monitoring system that classifies the fish's freshness based on body parts requires an object detection model, such as You Only Look Once (YOLO), for detecting the head and tail of fish. Yolov4-tiny is the recent tiny version of YOLO that has the advantage of a straightforward and fast in detecting objects. However, Yolov4-tiny obtain lower performance in object detection since the lack of diverse features generated by the backbone from the consecutive convolution layer. This paper proposes modifying Yolov4-tiny by inserting Spatial Pyramid Pooling (SPP) to expand the variety of feature maps using various kernel pooling from the same convolution layer. Our experimental results show that SPP increases the Recall of the model in detecting the expected object up to 83.42% and Recall up to 68.51% compared to the original versions.

Item Type:	Article
Uncontrolled Keywords:	yolov4-tiny, head, tail, object detection, spatial pyramid pooling
Subjects:	Technology
Divisions:	Faculty of Engineering
Depositing User:	Perpus Ubhara Surabaya
Date Deposited:	25 Jun 2024 03:35
Last Modified:	25 Jun 2024 03:35
URI:	http://eprints.ubhara.ac.id/id/eprint/2556

Actions (login required)

View Item