Title Page
Contents
ABSTRACT 15
Chapter 1. Introduction 17
1.1. Introduction 17
1.2. Background and Motivation 20
1.3. Problem Statement and Research Objectives 22
1.4. Overview of the Thesis Structure 23
Chapter 2. Literature Review 24
2.1. Deep Learning Predictions for Smart Farm Conditions 24
2.2. Computer Vision in Smart Farm Agriculture 25
2.2.1. Object Detection Applications in Smart Farm Agriculture 26
2.2.2. Crop Diseases Detection 28
2.2.3. Pests Detection 30
2.2.4. Plants Detection 31
2.2.5. Crop Detection 32
2.3. IoT Technologies in Smart Farm Agriculture 34
2.4. Autonomous Control Systems in Smart Farm Agriculture 38
2.5. Summary 39
Chapter 3. Methodology 40
3.1. AI Predictions for Smart Farm Conditions 40
3.1.1. Deep Neural Network for Environmental Predictions 41
3.1.2. Long-Short Term Memory for Environmental Predictions 43
3.1.3. One Dimensional Convolutional Neural Network for Environmental Predictions 46
3.2. A Novel adaptive DL model for Environmental Predictions 48
3.2.1. 1D-CNN Block 48
3.2.2. Genetic Algorithm 55
1.1. IoT Technologies for Real-Time Data Collection 58
1.2. YOLOv5 Object Detection Model 61
3.3.1. Why YOLOv5 61
3.3.2. YOLOv5 Object Detection Model Network Architecture 65
3.3.3. Hyperparameter Optimization of YOLOv5 Model 70
3.3.4. Training a Custom YOLOv5 Model 72
1.3. Autonomous Control System Development 74
1.4. Web-Based GUI for Remote Monitoring and Control 79
Chapter 4. Experimental Evaluation and Results 85
4.1. Experimental Setup: Smart Farm Prototype 85
4.1.1. Control Board: Jetson Nano 87
4.1.2. Sensors and Tools Used in Smart Farm 93
4.2. Experimental Setup: Tomato Detection in Outdoor Smart Farm 95
4.3. Datasets Collection and Preparation 98
4.3.1. Environmental Parameters Dataset of Smart Farm 98
4.3.2. Tomato, Plant and Pest Datasets 101
4.3.3. Dataset Augmentation 109
4.4. Environmental Prediction of The Smart Farm 111
4.4.1. Performance Metrics 111
4.4.2. Performance of DL Models 112
4.4.3. Experimental Confirmation 126
4.5. Object Detection Results 133
4.5.1. Performance Metrics 133
4.5.2. Performance of Optimized YOLOv5 on Tomato Dataset 135
4.5.3. Performance of Optimized YOLOv5 on Pests and Plants Datasets 146
Chapter 5. Conclusion and Future Work 152
References 154
Table 1. The best hyperparameters obtained by GA. 57
Table 2. Comparison results of YOLOv5 and YOLOv8 64
Table 3. Default hyperparameters of YOLOv5 model. 71
Table 4. Jetson Nano board interface description 90
Table 5. Jetson Nano board specifications 91
Table 6. AI board comparison 92
Table 7. Partition of all datasets 100
Table 8. The specific parameters of the datasets in this study. 108
Table 9. Comparison of the DNN model performance at different time intervals. 116
Table 10. Comparison of the LSTM model performance at different time intervals. 117
Table 11. Comparison of the 1D-CNN model performance at different time intervals. 118
Table 12. Comparison of the proposed model performance at different time intervals. 119
Table 13. Optimized hyperparameters of the YOLOv5s model. 138
Table 14. A comprehensive overview of the performance metrics for various YOLOv5 variants and optimized YOLOv5s model. 140
Figure 1. Representation of the evaluation of agricultural practice 19
Figure 2. Illustration of an IoT-enabled and DL-based smart smart farm. 37
Figure 3. DNN architecture used in this study. 42
Figure 4. (a) Architecture of the LSTM model used in this study and (b) the structure of a single cell of LSTM. 45
Figure 5. 1D-CNN architecture used in this study. 47
Figure 6. A graphical representation of directly connected two convolutional layers. 51
Figure 7. Depth-wise dimension reduction based on 1x1 convolutions. 52
Figure 8. a) 1D-CNN block, b) overall network architecture. 53
Figure 9. Flowchart of the proposed adaptive DL model-based climate parameters prediction method in hydroponic smart farm. 54
Figure 10. IoT system architecture of proposed smart hydroponic smart farm. 60
Figure 11. YOLOv5 model. 68
Figure 12. YOLOv5 model architecture. 69
Figure 13. Overall architecture of the autonomous system of the proposed smart hydroponic smart farm. 77
Figure 14. Overall flowchart of the controlling system of the smart farm. 78
Figure 15. Home page of the proposed web monitoring application. 80
Figure 16. Environmental condition prediction page of the proposed web monitoring application. 81
Figure 17. Plant detection page of the proposed web monitoring application. 82
Figure 18. Insect detection page of the proposed web monitoring application. 83
Figure 19. Crop production prediction page of the proposed web monitoring application. 84
Figure 20. Prototype of IoT and AI-based autonomous hydroponic smart farm. 86
Figure 21. Jetson Nano on the control board. 89
Figure 22. a) Camera installation process in a smart farm, b) schematic illustration of camera installation. 97
Figure 23. Samples from the tomato dataset. 104
Figure 24. Samples from plant dataset. 105
Figure 25. Samples from pest dataset. 106
Figure 26. Labeling process for detection: a) tomato dataset, b) plant dataset, c) pest dataset. 107
Figure 27. Comparison of the SEP per time for a temperature prediction. 120
Figure 28. Comparison of the SEP per time for humidity prediction. 121
Figure 29. Comparison of the SEP per time for CO₂ prediction. 122
Figure 30. Temperature prediction results using the DL models per time steps, (a) 1-minute time step, (b) 5-minute time steps, (c) 10-minute time steps, (d)... 123
Figure 31. Humidity prediction results using DL models per time steps, (a) 1-minute time step, (b) 5-minute time steps, (c) 10-minute time steps, (d) 15-... 124
Figure 32. CO₂ prediction results using DL models per time steps, (a) 1-minute time step, (b) 5-minute time steps, (c) 10-minute time steps, (d) 15-... 125
Figure 33. Comparison of climate parameters of smart farm with proposed system and without proposed system. 129
Figure 34. Comparison of plants condition of smart farm with proposed system and without proposed system. 130
Figure 35. Comparison of climate parameters of smart farm with proposed system and without proposed system for tomato plants. 131
Figure 36. Comparison of tomato plants' condition of smart farm with proposed system and without proposed system. 132
Figure 37. Visualization of hyperparameters of the YOLOv5s model. 139
Figure 38. Comparison of standard and optimized YOLOv5s detection model. 141
Figure 39. Identifying tomatoes under several challenging conditions. 142
Figure 40. Performance metrics of optimized model. 143
Figure 41. Tomato detection results in the outdoor smart farm. 144
Figure 42. Crop production prediction page of the web application. 145
Figure 43. Insect detection results 148
Figure 44. Insect detection evaluation results 149
Figure 45. Plant detection results 150
Figure 46. Plant detection evaluation results 151