Preventing spread of the invasive spotted lanternfly via texture-based automated egg detection

The invasive spotted lanternfly ( Lycorma delicatula ) threatens U.S. agriculture, particularly grape and tree fruit production. Early detection of egg masses is critical for limiting spread, yet current surveillance relies heavily on manual inspection, which is labor-intensive and difficult to scale. The lanternfly spreads primarily through human-assisted transport pathways, including trains, trucks, and freight infrastructure, enabling long-distance dispersal of egg masses. Here, we present a proof-of-concept automated image classification pipeline for SLF egg mass detection based exclusively on spatial texture features. Using a curated laboratory image dataset and descriptors including Gray-Level Co-occurrence Matrix (GLCM), GLDS (Gray Level Difference Statistics), and Hu and Zernike moments, we implemented a feature filtering and selection strategy to construct an interpretable, low-dimensional model. The final image-level screening classifier, a support vector machine with a radial basis function kernel trained on 12 selected features, achieved a mean Matthews Correlation Coefficient (MCC) of 0.881 (SD 0.037) under 5-fold stratified cross-validation. Generalization performance was evaluated on a held-out test set using bootstrap resampling (1,000 iterations), yielding a mean MCC of 0.836 (SD 0.037; 95% CI: 0.761–0.904). This image-level proof-of-concept under controlled imaging demonstrates that low-cost, scalable, and interpretable texture-based computer vision approaches may provide reliable early detection of SLF egg masses, supporting human-in-the-loop surveillance efforts in high-risk transport corridors and improving cost and reliability over manual inspection workflows.