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.