# Deep-Learning-based Counting Methods, Datasets, and Applications in Agriculture -- A Review

**Authors:** Guy Farjon, Liu Huijun, Yael Edan  
**arXiv:** [2303.02632](https://arxiv.org/abs/2303.02632)  
**Date:** 2023-03-05  
**Summary page:** https://www.summarizepaper.com/en/paper/2303.02632v2/

## Key points

- Literature review on counting methods in agriculture, with a focus on deep learning approaches
- Five-step bibliometric analysis conducted to select papers for review
- Two databases (Web of Science and Scopus) used to search for peer-reviewed papers in agricultural domain
- Initial search identified 2290 papers, narrowed down to 282 after reading titles and abstracts
- Final selection consisted of 243 peer-reviewed papers (22 reviews, 221 articles)
- Evolution of methods for automating counting tasks in agriculture analyzed based on various factors
- Increasing number of papers incorporating counting components in agriculture over the years
- Review papers covered topics such as fruit yield estimation, insect pest monitoring, animal and crop monitoring, phenotyping
- Shift towards automatic data collection using robots or unmanned aerial vehicles (UAVs)
- Research expanded from studying single varieties at specific times/places to multi-variety systems across seasons/geographical locations
- Advanced sensors used to capture more detailed information about agricultural features
- Progress made in object counting in agriculture due to adoption of deep learning methods

## Summary

This article presents a literature review on counting methods in agriculture, with a focus on deep learning approaches. The review was conducted using a five-step bibliometric analysis that included selecting databases, defining search criteria, screening papers, selecting final papers for thorough reading and analyzing the results. Two databases - Web of Science and Scopus - were used to search for peer-reviewed papers in the agricultural domain with a counting component. The initial search identified 2290 papers which were narrowed down to 282 after careful reading of titles and abstracts. The final selection consisted of 243 peer-reviewed papers including 22 reviews and 221 articles. The evolution of methods for automating counting tasks in agriculture was analyzed based on the type of objects, sensors, algorithmic approaches, application areas, platforms and environments as well as setting variations. The results revealed an increasing number of papers incorporating counting components in agriculture over the years. Among the review papers identified none specifically targeted object counting in agriculture but provided information on topics such as fruit yield estimation, insect pest monitoring, animal and crop monitoring and phenotyping. The use of different platforms for data collection was examined across the years. While manual data collection remains dominant through the use of tripods or manual image capture techniques there has been an increase in automatic data collection using robots or unmanned aerial vehicles (UAVs). This shift towards automatic data collection enables massive data collection but also introduces additional challenges. Research has also expanded from studying single varieties at specific times and places to considering multi-variety systems across various seasons and geographical locations. Advanced sensors are being used to capture more detailed information about agricultural features. In conclusion this literature review highlights the progress made in object counting in agriculture over the past decade due to adoption of deep learning methods. It provides insights into different aspects related to counting algorithms metrics platforms sensors datasets and challenges faced in this field.