Autonomous Multi-Model Vision System for Site-Specific Weed Detection and Selective Spraying in Smart Agriculture

Benchmarking YOLOv5 and MobileNet-SSD for Field-Deployable Crop-Weed Discrimination

DOI

10.2991/978-94-6239-821-4_2How to use a DOI?

Keywords

Precision agriculture; weed detection; YOLOv5; MobileNet-SSD; deep learning; robotic weeding; edge AI; smart farming

Abstract

The impact of weed infestation is a serious threat to global food security. Weeds can reduce crop yield by as much as 43%. Mechanical, manual, and chemical methods have been widely used to control weeds. However, the rising costs of labor, the development of herbicide resistance in weeds, and restrictions on chemical warfare are significant concerns for modern agriculture. This article introduces an autonomous, multimodal robotic solution that scans its environment to detect and remove weeds. Initially, we used 3,120 images of crops, weeds, and backgrounds extracted from Internet resources and field photos. These images were annotated, oriented, and resized to 416 × 416 pixels. Moreover, we applied flips and increased the contrast and exposure to obtain an extended data set. We trained and compared two popular object detectors, including YOLOv5 and MobileNet-SSD, to find the optimal balance between performance and a compact model for further deployment. The mean Average Precision (mAP) of YOLOv5 at 0.5 was approximately 0.86, while the maximum F1-score was 0.83 at a confidence threshold of around 0.49. MobileNet-SSD as a classifier has demonstrated strong performance with an accuracy of 98%. As a result, we also designed and built a mechanical chassis for the robot. Thus, with the obtained results and the analysis of the comparison between the two approaches, we prioritize the following steps for prototype development and field testing.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Sana I. Shaikh
AU  - Mayur Jakhete
PY  - 2026
DA  - 2026/10/31
TI  - Autonomous Multi-Model Vision System for Site-Specific Weed Detection and Selective Spraying in Smart Agriculture
BT  - Proceedings of the International Conference on Information Systems Design And Intelligent Applications (ICISDA 2026)
PB  - Atlantis Press
SP  - 4
EP  - 13
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-821-4_2
DO  - 10.2991/978-94-6239-821-4_2
ID  - Shaikh2026
ER  -