Enhancing Agricultural Pest Detection Using Machine Learning and Deep Learning

[Virtual Presenter] Good evening everyone, We are here today to present our research on enhancing agricultural pest detection using machine learning and deep learning. We believe that this new approach will provide a more reliable and cost-effective way to detect pests, compared to traditional methods. We hope this presentation will help you to have a better understanding of the potential of this approach. As we move forward, we will discuss the dataset and the methodology we used, the key findings we achieved and the next steps for implementation. We appreciate your time and look forward to the discussion..

[Audio] Agriculture plays an important role in our global economy and society. It supplies nutrition for humans and additional jobs in rural areas, as well as industries such as pharmaceuticals and textiles. To ensure its continued success, it is essential to continue to cultivate the development of agriculture. To do so, we must take advantage of cutting-edge technologies, including machine learning and deep learning, to identify and address agricultural threats..

[Audio] Pests in agriculture can have a devastating effect on food security and sustainable farming practices worldwide. Current methods of pest detection are difficult, lengthy, and unreliable, causing financial losses for farmers and agricultural systems. This conference is hoping to leverage the potential of machine and deep learning to better detect pests, and to safeguard agriculture from their damage over time. Through pioneering solutions, we have the power to ensure agricultural resilience for the future..

[Audio] There is a significant amount of work done in the field of agricultural pest detection using machine learning and deep learning. For example, Xin et al. employed a DCNN model for recognizing crop diseases and insect pests, while Wang et al. focused on deep CNN for crop pest image classification and Thenmozhi et al. explored transfer learning for crop pest classification. Wang et al. also used DCNN networks for rapid pest recognition in agriculture and forestry, and Venugoban et al. concentrated on paddy field insects, utilizing gradient-based features for classification. Lastly, Li. et al. applied residual neural networks and transfer learning for pest image classification. This conference seeks to further progress the research conducted by these authors..

[Audio] A comprehensive dataset for agricultural pest detection and categorization is proposed in this study. The dataset, sourced from Kaggle and curated through the Flickr API, includes 12 different pest categories. The images are resized in order to build a robust machine learning model. The variety of images in the dataset allows for the construction of a reliable model that can detect essential features like shape and color. The distribution of images across the categories is illustrated in Figure 2..

[Audio] Focusing on using machine learning and deep learning to detect agricultural pests, we used the Kaggle repository to access datasets across 12 pest categories. OpenCV was used to preprocess and scale the images and apply data augmentation techniques, such as rotation, flipping, cropping, and noise injection, to diversify the training dataset. To identify the most informative attributes from the dataset, Chi2 selection and SelectKBest with k=500 was implemented to balance the feature dimensionality without compromising accuracy..

[Audio] The dataset collection and preprocessing processes are integral to the agricultural pest detection process. Various deep learning and machine learning models have been employed to extract features and classify them. Each model was thoroughly evaluated for performance, and the best performing deep learning and machine learning models were identified. A hybrid model was then implemented to further improve accuracy..

[Audio] We will demonstrate our process for agricultural pest detection using machine learning and deep learning in this slide. Our dataset collection will be presented first, then we will discuss our preprocessing steps. Afterwards, feature extraction using a Convolutional Neural Network (CNN) will be done, and classification using machine learning algorithms will be carried out last. Following this process will enable us to effectively detect agricultural pests with high accuracy..

[Audio] Cloud platforms played a pivotal role in this conference. We took advantage of Google Colab and Kaggle Notebooks to facilitate teamwork and exploit cloud platform features to their fullest potential. With Google Colab Notebooks, we accessed TPUs or T4 GPUs, and Kaggle Notebooks provided a wider range of powerful solutions including 2x T4 GPUs, P100 GPUs, or TPU VM v3-8. We then scrutinized the framework and architecture of ML and DL models to determine the ideal and most efficient hybrid model..

[Audio] The performance metrics of Accuracy, Precision, and F1-Score are used to ensure the model's effectiveness in classifying agricultural pests. Our analysis of the results further confirms that this model can correctly and reliably be used for detecting agricultural pests..

COMPARISION OF RESULTS OF ALL MODELS. ACCURACY. PRECISION.

[Audio] We have created a hybrid model that merges machine learning and deep learning models. Our ML model merely analyzed a small number of image attributes, while our DL model considered a number of other elements, with an accuracy rate of 96.7%. We have observed a major enhancement in our classification output and in our ML model training. Nevertheless, there are still possibilities to enhance by searching for solutions to reduce the cost of running deep learning models..

[Audio] Our research has shown the great potential in combining Deep Learning and Machine Learning models to accurately detect agricultural pests, reaching an accuracy rate of 96.7%. By integrating EfficientNetB0 and XGBoostClassifier, our model enables farmers to detect pests in a timely manner, optimizing their procedures and enhancing efficiency. This opens avenues of possibilities for future research, motivating further development of the model with more pictures of pests in various classes..

Thank you!.