FE - DEM | Documentos por Auto-Depósito
Permanent URI for this collection
Browse
Browsing FE - DEM | Documentos por Auto-Depósito by advisor "Lima, Tânia Daniela Felgueiras de Miranda"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- Application of Artificial Intelligence algorithms to support decision-making in agriculture activitiesPublication . Alibabaei, Khadijeh; Gaspar, Pedro Miguel de Figueiredo Dinis Oliveira; Lima, Tânia Daniela Felgueiras de MirandaDeep Learning has been successfully applied to image recognition, speech recognition, and natural language processing in recent years. Therefore, there has been an incentive to apply it in other fields as well. The field of agriculture is one of the most important in which the application of artificial intelligence algorithms, and particularly, of deep learning needs to be explored, as it has a direct impact on human well-being. In particular, there is a need to explore how deep learning models for decision-making can be used as a tool for optimal planting, land use, yield improvement, production/disease/pest control, and other activities. The vast amount of data received from sensors in smart farms makes it possible to use deep learning as a model for decision-making in this field. In agriculture, no two environments are exactly alike, which makes testing, validating, and successfully implementing such technologies much more complex than in most other sectors. Recent scientific developments in the field of deep learning, applied to agriculture, are reviewed and some challenges and potential solutions using deep learning algorithms in agriculture are discussed. Higher performance in terms of accuracy and lower inference time can be achieved, and the models can be made useful in real-world applications. Finally, some opportunities for future research in this area are suggested. The ability of artificial neural networks, specifically Long Short-Term Memory (LSTM) and Bidirectional LSTM (BLSTM), to model daily reference evapotranspiration and soil water content is investigated. The application of these techniques to predict these parameters was tested for three sites in Portugal. A single-layer BLSTM with 512 nodes was selected. Bayesian optimization was used to determine the hyperparameters, such as learning rate, decay, batch size, and dropout size. The model achieved mean square error (MSE) values ranging from 0.07 to 0.27 (mm d–1)² for ETo (Reference Evapotranspiration) and 0.014 to 0.056 (m³m–3)² for SWC (Soil Water Content), with R2 values ranging from 0.96 to 0.98. A Convolutional Neural Network (CNN) model was added to the LSTM to investigate potential performance improvement. Performance dropped in all datasets due to the complexity of the model. The performance of the models was also compared with CNN, traditional machine learning algorithms Support Vector Regression, and Random Forest. LSTM achieved the best performance. Finally, the impact of the loss function on the performance of the proposed models was investigated. The model with the mean square error (MSE) as loss function performed better than the model with other loss functions. Afterwards, the capabilities of these models and their extension, BLSTM and Bidirectional Gated Recurrent Units (BGRU) to predict end-of-season yields are investigated. The models use historical data, including climate data, irrigation scheduling, and soil water content, to estimate endof- season yield. The application of this technique was tested for tomato and potato yields at a site in Portugal. The BLSTM network outperformed the GRU, the LSTM, and the BGRU networks on the validation dataset. The model was able to capture the nonlinear relationship between irrigation amount, climate data, and soil water content and predict yield with an MSE of 0.017 to 0.039 kg/ha. The performance of the BLSTM in the test was compared with the most commonly used deep learning method called CNN, and machine learning methods including a Multi-Layer Perceptrons model and Random Forest regression. The BLSTM out-performed the other models with a R2-score between 0.97 and 0.99. The results show that analyzing agricultural data with the LSTM model improves the performance of the model in terms of accuracy. The CNN model achieved the second-best performance. Therefore, the deep learning model has a remarkable ability to predict the yield at the end of the season. Additionally, a Deep Q-Network was trained for irrigation scheduling. The agent was trained to schedule irrigation for a tomato field in Portugal. Two LSTM models trained previously were used as the agent environment. One predicts the total water in the soil profile on the next day. The other one was employed to estimate the yield based on the environmental condition during a season and then measure the net return. The agent uses this information to decide the following irrigation amount. LSTM and CNN networks were used to estimate the Q-table during training. Unlike the LSTM model, the ANN and the CNN could not estimate the Qtable, and the agent’s reward decreased during training. The comparison of the performance of the model was done with fixed-base irrigation and threshold-based irrigation. The trained model increased productivity by 11% and decreased water consumption by 20% to 30% compared to the fixed method. Also, an on-policy model, Advantage Actor–Critic (A2C), was implemented to compare irrigation scheduling with Deep Q-Network for the same tomato crop. The results show that the on-policy model A2C reduced water consumption by 20% compared to Deep Q-Network with a slight change in the net reward. These models can be developed to be applied to other cultures with high importance in Portugal, such as fruit, cereals, and grapevines, which also have large water requirements. The models developed along this thesis can be re-evaluated and trained with historical data from other cultures with high production in Portugal, such as fruits, cereals, and grapes, which also have high water demand, to create a decision support and recommendation system that tells farmers when and how much to irrigate. This system helps farmers avoid wasting water without reducing productivity. This thesis aims to contribute to the future steps in the development of precision agriculture and agricultural robotics. The models developed in this thesis are relevant to support decision-making in agricultural activities, aimed at optimizing resources, reducing time and costs, and maximizing production.