RESNET-50 Based Pest Identification in Plants

The advent of deep learning in smart agriculture has notably enhanced the sector’s capabilities. However, the performance of these advanced systems is frequently constrained by issues related to unreliable data annotation and poor data quality. To mitigate these challenges, we introduced an innovative method using distance-entropy to distinguish between highquality and low-quality data from an informational perspective. Our approach leveraged ResNet-50 as the core network and involved extensive trials with the IP06 dataset. The results uncovered a significant amount of redundant data, with nearly $50 \%$ of the dataset containing enough duplicates to achieve similar accuracy in testing. Through an in-depth analysis of these representative samples, we devised recommendations to improve dataset efficiency. These guidelines offer a strategic roadmap for enhancing data-driven research in smart agriculture, promoting innovation and sustainability by optimizing data use.