An Efficient Hybrid Machine Learning and Deep Learning-Base on Agricultural Pest Detection System

Abhishek Gadilkar

Abstract: The attack of agricultural pests is still considered a significant problem that adversely affects crop production and food security worldwide. The conventional techniques used to detect agricultural pests rely on manual visual inspection and are time-consuming, inefficient, and imprecise. This research proposes a sophisticated pest detection approach that uses machine and deep learning techniques. Traditional techniques used in pest detection include manual inspection and normal farming practices, both of which tend to consume lots of time and effort. Moreover, such techniques have a higher chance of failing since they cannot detect pests accurately. Therefore, this research aims to develop an innovative pest detection system through the implementation of a combination of ML and DL technologies. For this purpose, CNNs will be employed in extracting characteristics from pests' images without the need for feature extraction. Such techniques involve the use of advanced architectural designs, such as EfficientNet, ResNet, and Inception. Then, the extracted features are processed by ML algorithms, including SVM and XGBoost. Thus, the use of the hybrid model in pest classification enhances accuracy and efficiency. In addition, the application of image preprocessing and augmentation techniques will contribute significantly towards improving the accuracy of pest recognition systems. The evaluation of the developed model will be performed based-score criteria. Besides that, the effectiveness of the proposed pest detection technology will be proved by the ability of the model to provide accurate results in real-time. In addition, such an intelligent pest recognition model could be incorporated into user-friendly web interfaces in order to enable farmers to upload images and receive accurate pest detection results instantly.

Keywords: agricultural pest detection, machine learning, deep learning, image-based pest recognition, real-time crop monitoring