Accurate identification within the Tenebrionidae triad—composed of Tribolium castaneum, Tribolium confusum, and Latheticus oryzae—is essential for improving stored product pest management. These beetles exhibit subtle and overlapping morphological traits that often challenge traditional identification techniques. This study presents an engineering-based image analysis system for classifying species using anatomical fragment images, including the elytra, pronotum, antennae, and head aspect ratio regions. The head aspect ratio fragment captures proportional measurements between the frons and vertex, offering an additional distinguishing feature among morphologically similar species. The image dataset was processed using a convolutional neural network architecture based on EfficientNet B0. Selected for its balanced accuracy and model efficiency, EfficientNet B0 enables detailed extraction of fine morphological structures while remaining suitable for low-power deployment environments. To enhance feature localization, heatmaps were generated and annotated across each anatomical fragment, highlighting regions of interest that contributed most significantly to classification. The trained model achieved a classification accuracy of 94.7% across the three species. This approach enhances species-level identification capabilities by integrating imaging precision with machine learning, contributing to practical applications in automated pest surveillance systems for postharvest storage environments. The outcome of this study holds substantial value not only for researchers and entomologists but also for farmers, by offering a reliable and scalable tool to distinguish between visually similar beetle pests that impact stored grain quality and safety.
.png)
