RT info:eu-repo/semantics/article
T1 Simplifying YOLOv5 for deployment in a real crop monitoring setting
A1 Nnadozie, Emmanuel Chibuikem
A1 Casaseca de la Higuera, Juan Pablo
A1 Iloanusi, Ogechukwu
A1 Ani, Ozoemena
A1 Alberola López, Carlos
K1 Object detection
K1 Model simplification
K1 Crop monitoring
K1 YOLOv5
K1 Deep learning
K1 33 Ciencias Tecnológicas
AB Deep learning-based object detection models have become a preferred choice for cropdetection tasks in crop monitoring activities due to their high accuracy and generalizationcapabilities. However, their high computational demand and large memory footprint pose achallenge for use on mobile embedded devices deployed in crop monitoring settings. Vari-ous approaches have been taken to minimize the computational cost and reduce the size ofobject detection models such as channel and layer pruning, detection head searching, back-bone optimization, etc. In this work, we approached computational lightening, model com-pression, and speed improvement by discarding one or more of the three detection scalesof the YOLOv5 object detection model. Thus, we derived up to five separate fast and lightmodels, each with only one or two detection scales. To evaluate the new models for a realcrop monitoring use case, the models were deployed on NVIDIA Jetson nano and NVIDIAJetson Orin devices. The new models achieved up to 21.4% reduction in giga floating-pointoperations per second (GFLOPS), 31.9% reduction in number of parameters, 30.8% reduc-tion in model size, 28.1% increase in inference speed, with only a small average accuracydrop of 3.6%. These new models are suitable for crop detection tasks since the crops areusually of similar sizes due to the high likelihood of being in the same growth stage, thus,making it sufficient to detect the crops with just one or two detection scales.
PB Springer
SN 1380-7501
YR 2023
FD 2023
LK https://uvadoc.uva.es/handle/10324/66697
UL https://uvadoc.uva.es/handle/10324/66697
LA eng
NO Multimedia Tools and Applications, 2023.
NO Producción Científica
DS UVaDOC
RD 21-nov-2024