Enhancing Agricultural Pest Detection Using Machine Learning and Deep Learning

2ND INTERNATIONAL CONFERENCE ON AMBIENT INTELLIGENCE IN HEALTH CARE.

INTRODUCTION. Agriculture's Crucial Role: Agriculture serves as the backbone of economic development and human nutrition. It provides a consistent and reliable food source, stimulating economic growth and job creation, especially in rural areas..

NEED OF PEST DETECTION. The research is motivated by the critical need to address persistent threats from agricultural pests, which jeopardize global food security and sustainable farming practices. These pests pose significant challenges, including the degradation of crop production, diminished food quality, and heightened economic burdens on farmers and agricultural systems worldwide. The urgency for novel and effective pest detection systems is underscored by the increasing global population and the exacerbating impact of climate change on pest-related difficulties. Issues with Traditional Methods: Reliance on physical labor and visual examination. Must meet accuracy, scalability, and speed expectations. Time-consuming, leading to significant crop losses..

AVAILAIBLE WORK. Xin et al. [1] employed a DCNN model for recognizing crop diseases and insect pests. Wang et al. [2] focused on deep CNN for crop pest image classification. Thenmozhi et al. [3] explored transfer learning for crop pest classification. Wang et al. [4] used DCNN networks for rapid pest recognition in agriculture and forestry. Venugoban et al. [5] concentrated on paddy field insects, utilizing gradient-based features for classification. Li. et al. [6] applied residual neural networks and transfer learning for pest image classification..

PROPOSED WORK. The study utilizes a dataset from Kaggle, encompassing 12 diverse pest categories, curated through the Flickr API. Resized for manageability, the dataset captures authentic scenarios, incorporating essential attributes like shape and color. This comprehensive approach enhances the dataset's potential for developing a robust model for agricultural pest detection and categorization. Figure 2 illustrates the distribution of images across all 12 classes, providing insights into the dataset's diversity..

METHODS. Utilized Kaggle repository for a diverse dataset across 12 pest categories. Images organized into labeled folders for pest-type association. OpenCV used for scaling, normalization, and other preprocessing tasks. Applied data augmentation techniques (rotation, flipping, cropping, noise injection) for diversifying the training dataset..

Dataset Collection. METHODS. Preprocessing. Deep Learning Models.

Dataset Collection. WORKFLOW. Preprocessing. Feature Extraction Using CNN.

RESULT ANALYSIS. Leveraged Google Colab [14] and Kaggle Notebooks [15] for real-time collaboration and utilizing cloud platform capabilities. Google Colab Notebooks with embedded "TPU" or "T4 GPU" for accelerated research. Kaggle Notebooks with options for "2 x T4 GPU," "P100 GPU," or "TPU VM v3-8.".

RESULT ANALYSIS. Comprehensive evaluation metrics considered: Accuracy (ACC): Overall correctness of predictions. Precision (P): Proportion of true positive predictions among all positive predictions. F1-Score: Harmonizes precision and sensitivity for a balanced performance measure..

COMPARISION OF RESULTS OF ALL MODELS. ACCURACY. PRECISION.

DISCUSSION. Noteworthy outcome: Development of a Hybrid Model combining DL and ML models. Achieved an excellent accuracy rate of 96.7%..

CONCLUSION. Our research demonstrates that combining Deep Learning (DL) and Machine Learning (ML) models outperforms using either alone, achieving a remarkable 96.7% accuracy. The hybrid model, integrating EfficientNetB0 and XGBoostClassifier, allows farmers to detect pests early, streamlining their tasks. Additionally, our findings open avenues for future research, encouraging the expansion of the model with more pest photos from various classes to enhance hybrid learning in agricultural pest detection..

Thank you!.