Detection of Bemisia tabaci based on SwinIR super-resolution reconstruction and semantic-SAM model
Year: 2025
Type: article
Abstract: Bemisia tabaci is a common pest on tobacco, which seriously endangers the yield and quality of tobacco leaves. Detection of B. tabaci and accurate calculation of insect population density are the basis for formulating prevention and control strategies. This paper proposes a detection method for whitefly based on SwinIR super-resolution reconstruction and Semantic-SAM model, which can detect and classify B. tabaci with different development stages on tobacco leaves. Firstly, images of B. tabaci on tobacco leaves were collected, and a data set of B. tabaci images was established. Secondly, the SwinIR (Image Restoration Using Swin Transformer) model was used for image super-resolution reconstruction to obtain detailed and clear images. In the B. tabaci identification stage, the Semantic-SAM (Semantic-Segment Anything Model) model was introduced to initially classify B. tabaci ; combined with the feret diameter, the instars of B. tabaci were assisted to distinguish, thus strengthening the classification effect. The experimental results showed that the average recognition accuracy rate of adults was the highest, reaching 100 %; the average recognition rates of the second and third instar nymphs were 86 % and 88 %, respectively; the average recognition rates of eggs, first and fourth instar nymphs were 90 %. The method of combining the SwinIR model, Semantic-SAM model and feret diameter detection proposed in this study can realize the analysis of the development stages and quantity of B. tabaci on tobacco leaves, which is of great significance for the prediction and control of pest populations.
Language: English
Topic: Smart Agriculture and AI
Subfield: Plant Science
Domain: Life Sciences
SDG: Zero hunger
Open Access status: hybrid
Estimated APC paid: $3,870
OpenAlex ID: W4411333342