Papers
Topics
Authors
Recent
Search
2000 character limit reached

Deep learning in agriculture: A survey

Published 31 Jul 2018 in cs.LG, cs.CV, and stat.ML | (1807.11809v1)

Abstract: Deep learning constitutes a recent, modern technique for image processing and data analysis, with promising results and large potential. As deep learning has been successfully applied in various domains, it has recently entered also the domain of agriculture. In this paper, we perform a survey of 40 research efforts that employ deep learning techniques, applied to various agricultural and food production challenges. We examine the particular agricultural problems under study, the specific models and frameworks employed, the sources, nature and pre-processing of data used, and the overall performance achieved according to the metrics used at each work under study. Moreover, we study comparisons of deep learning with other existing popular techniques, in respect to differences in classification or regression performance. Our findings indicate that deep learning provides high accuracy, outperforming existing commonly used image processing techniques.

Citations (2,678)

Summary

  • The paper presents a systematic review of 40 studies, analyzing deep learning models like CNNs and RNNs applied to agricultural challenges.
  • The paper demonstrates significant performance gains, with DL-based applications often achieving classification accuracies above 90% compared to traditional methods.
  • The paper identifies key challenges such as data scarcity and annotation complexity, while advocating for future innovations in smart farming and automated systems.

Deep Learning in Agriculture: A Survey

The paper "Deep Learning in Agriculture: A Survey" by Andreas Kamilaris and Francesc X. Prenafeta-Boldú provides a comprehensive survey of the application of deep learning (DL) techniques to various challenges in agriculture and food production. The review encompasses 40 research efforts, scrutinizing the agricultural problems addressed, models and frameworks employed, data sources and preprocessing methods used, and the performance metrics adopted. The survey distinctly highlights comparisons of DL models with other existing popular methodologies, showcasing substantial improvements obtained via DL.

Overview of Deep Learning Applications

Deep learning, specifically Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN), has shown high precision in image processing tasks within agriculture. With DL's capability for automatic feature extraction and hierarchical data representation, researchers have demonstrated notable improvements in agricultural image classification, anomaly detection, and regression tasks. This survey categorizes the diversity of agricultural challenges tackled using DL into areas such as fruit counting, crop and weed detection, land cover classification, and plant disease identification.

Data Sources and Preprocessing

The studies generally utilize large datasets, some publicly available and others collected by researchers. Common sources include imaging datasets such as PlantVillage, LifeCLEF, and Flavia, along with remote sensing platforms like satellites, UAVs, and hyperspectral cameras. Preprocessing steps are critical for enhancing the quality and relevancy of the data. Techniques such as image resizing, segmentation, background removal, and the use of vegetation indices like NDVI are widespread practice. Data augmentation through transformations (rotations, scaling, cropping) is also frequently employed to address the scarcity of data and improve model generalization.

Technical Implementation and Performance

The survey observes frequent use of well-known CNN architectures such as AlexNet, VGG16, and Inception-ResNet. Approximately 42% of the studies employed these architectures, with frameworks like Caffe and TensorFlow being prevalent choices for implementation. Transfer learning—a technique where pre-trained models on generic datasets are fine-tuned for specific agricultural tasks—proved effective in boosting performance metrics without the necessity to train from scratch.

Performance metrics varied widely, with Classification Accuracy (CA), F1 score, Root Mean Square Error (RMSE), and Intersection over Union (IoU) being predominant. Notably, a significant portion of the studies reported high CA values above 90%, emphasizing DL's superior performance over traditional methods like SVM, RF, and ANN. Specifically, studies like Mohanty et al. (2016) and Lee et al. (2015) reported CA values exceeding 98%.

Comparisons with Traditional Techniques

A recurring theme in this survey is the superior ability of DL models to outperform other techniques across numerous applications. In 70% of the papers that included direct comparisons, DL models demonstrated better performance. For instance, CNNs showed 1-8% higher CA compared to SVM and 3-8% improvement over RF models. This marked advantage is largely attributed to DL's proficiency in automatic feature extraction and managing complex data representations.

Challenges and Limitations

Despite its advantages, DL in agriculture faces challenges such as the need for large and well-annotated datasets. Manual annotations require experts and are time-consuming, making scalability a significant hurdle. Issues like low variation among classes, occlusion, and noisy data further complicate model training and generalization. Additionally, while DL models excel in specific tasks, their requirement for extensive computational resources and longer training times can be limiting factors.

Future Directions and Implications

The potential applications of DL in agriculture are extensive, especially with advancements in RNNs for temporal data analysis and the expected emergence of more complex DL architectures. Future research could explore untested areas such as pest detection, water stress assessment, and food contamination identification using DL. Moreover, the commercial deployment of DL models in autonomous agricultural robots and real-time decision support systems could revolutionize smart farming practices.

The authors suggest that the continued development of pre-trained models, publicly available datasets, and efficient frameworks will be pivotal in progressing DL applications in agriculture. Increased collaboration and data sharing within the research community can drive innovation, leading to more robust and generalized agricultural DL solutions.

Conclusion

This survey underscores the promising role of deep learning in addressing critical agricultural challenges, citing superior performance and extensive applicability across diverse tasks. While acknowledging existing limitations and the need for further exploration, the survey sets a solid foundation for future research endeavors aimed at integrating DL into sustainable and intelligent agricultural systems.

Paper to Video (Beta)

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Collections

Sign up for free to add this paper to one or more collections.