| Andreas Kamilaris and Francesc X. Prenafeta-Boldú
This paper provides a comprehensive survey of 40 research efforts that apply deep learning (DL) techniques to various agricultural and food production challenges. The authors examine the specific agricultural problems addressed, the models and frameworks used, the data sources and preprocessing methods, and the performance metrics achieved. They also compare the performance of DL with other popular techniques in terms of classification and regression accuracy. The findings indicate that DL outperforms existing image processing techniques, achieving high accuracy in tasks such as weed detection, land cover classification, plant recognition, fruit counting, and crop type classification. The survey highlights the advantages of DL, including reduced feature engineering effort and improved generalization capabilities, while discussing its disadvantages such as longer training times and potential issues with small or diverse datasets. The paper concludes by discussing advanced DL applications and the potential of RNN-based models for capturing temporal dependencies in agricultural data.This paper provides a comprehensive survey of 40 research efforts that apply deep learning (DL) techniques to various agricultural and food production challenges. The authors examine the specific agricultural problems addressed, the models and frameworks used, the data sources and preprocessing methods, and the performance metrics achieved. They also compare the performance of DL with other popular techniques in terms of classification and regression accuracy. The findings indicate that DL outperforms existing image processing techniques, achieving high accuracy in tasks such as weed detection, land cover classification, plant recognition, fruit counting, and crop type classification. The survey highlights the advantages of DL, including reduced feature engineering effort and improved generalization capabilities, while discussing its disadvantages such as longer training times and potential issues with small or diverse datasets. The paper concludes by discussing advanced DL applications and the potential of RNN-based models for capturing temporal dependencies in agricultural data.