# Deep learning in agriculture: A survey

**Authors:** Andreas Kamilaris, Francesc X. Prenafeta-Boldu  
**arXiv:** [1807.11809](https://arxiv.org/abs/1807.11809)  
**Date:** 2018-07-31  
**Summary page:** https://www.summarizepaper.com/en/paper/1807.11809v1/

## Key points

- Deep learning is a powerful technique for image processing and data analysis
- It has shown promising results and great potential in various domains, including agriculture
- A comprehensive survey of 40 research efforts utilizing deep learning in agriculture was conducted
- The survey examined specific agricultural problems studied, models and frameworks used, data sources and preprocessing techniques applied
- Performance of each work was evaluated based on specific metrics and compared to other techniques
- Large datasets consisting of thousands of images are commonly used, but smaller sets from UAVs, airborne platforms, remote sensing systems or X-ray tomography are also employed for certain problems
- The amount of data required for training deep learning models depends on the complexity of the problem being addressed
- Deep learning extends traditional machine learning methods by incorporating more complexity into models and allowing for automatic feature extraction from raw data without manual intervention
- Overall, deep learning offers high accuracy when addressing various agricultural challenges

## Summary

Deep learning has emerged as a powerful technique for image processing and data analysis, showing promising results and great potential in various domains. Recently, it has also made its way into the field of agriculture. In this paper, the authors conduct a comprehensive survey of 40 research efforts that utilize deep learning techniques to address agricultural and food production challenges. The authors examine the specific agricultural problems studied in each research effort, the models and frameworks employed, the sources and nature of data used, as well as the preprocessing techniques applied. They also evaluate the overall performance achieved by each work based on specific metrics. Additionally, comparisons between deep learning and other popular techniques are analyzed in terms of classification or regression performance. One key aspect explored is the data sources used to train deep learning models. The majority of papers rely on large datasets consisting of thousands of images. These datasets can either be real images collected by researchers or synthetic ones generated specifically for their studies. Well-known publicly available datasets such as PlantVillage, LifeCLEF, MalayaKew, UC Merced, and Flavia are commonly utilized. However, some papers dealing with land cover classification, crop type identification, yield estimation, and weed detection employ smaller sets of tens or hundreds of images obtained from UAVs (Unmanned Aerial Vehicles), airborne platforms, satellite-based remote sensing systems or even X-ray tomography. The complexity of the problem being addressed often determines the amount of data required for training deep learning models. For instance, problems involving a large number of classes or small variations among classes necessitate a larger number of input images to achieve accurate results. Furthermore, the authors investigate whether researchers compared their deep learning approach with other techniques and if there were any differences in performance observed. The detailed findings are presented in Section 4 and can be found in Appendix II. Deep learning extends traditional machine learning methods by incorporating more complexity into models and employing various functions to transform data hierarchically. This allows for automatic feature extraction from raw data without manual intervention which leads to higher accuracy than conventional image processing techniques due to its ability to parallelize computations effectively.. Overall ,the findings from this survey indicate that deep learning offers high accuracy when addressing various agricultural challenges providing valuable insights for future research in this field .