Deep learning based agricultural pest monitoring and classification

Precise pest classification plays an essential role in smart agriculture. Crop yields are severely impacted by pest damage, which poses a critical challenge for agricultural production and the economy. Identifying pests is of utmost importance, but manual identification is both labor-intensive and time-consuming. Therefore, the realm of pest identification and classification requires more advanced and effective techniques. The proposed work presents an innovative automatic approach based on the incorporation of deep learning in smart farming for pest monitoring and classification to tackle this challenge. In this work, the IP102 dataset is used to identify and classify 82 classes of pests. Autoencoder is utilized to address data imbalance issue by generating augmented images. RedGreenBlue colour code and object detection techniques are employed to localize and segment pests from the field images. Finally, these segmented pests are classified using Convolutional neural networks. The Average Intersection of Union (IoU) of object detection used for pest segmentation is 80%. The proposed classification model achieved an accuracy of 84.95% with the balanced dataset, outperforming the existing model. Identifying the count of pests in the image helps in determining the extent of pest damage. The results showcase the potential of this approach to revolutionize traditional pest monitoring methods, offering a more proactive and precise strategy for pest control in agricultural settings. This research work contributes to the advancement of smart farming practices through intelligent pest classification for pest control.