InsectMorphoAI: A deep learning framework for automated insect morphometrics and biomass estimation with taxon-specific volumetric validation

Accurate measurement of insect morphometric traits is essential for functional ecology and biodiversity monitoring. Yet traditional manual methods are labor-intensive, invasive, destructive, and difficult to scale within high-throughput biodiversity pipelines. InsectMorphoAI is an open-source, dual-module framework that automates specimen-level trait extraction from standard 2D images using deep learning. A hardware-agnostic Oriented Bounding Box (OBB) module provides rotation-invariant linear length estimation across diverse insect taxa, achieving a mean absolute error of 0.211 mm (≈2.3% of mean specimen length) in an independent validation dataset of bristle flies (Diptera: Tachinidae). A complementary instance segmentation module enables high-precision, taxon-specific trait extraction by delineating the head, thorax, and abdomen to derive curvilinear lengths and approximate 3D body volume using stacked frustums. Implemented and validated here using bristle flies as a proof-of-concept system, segmentation-derived volume explained more variation in body-only biomass than linear length within the validation dataset (R 2 = 0.823 for body-only dry weight and R 2 = 0.880 for wet weight after leg removal, compared to R 2 = 0.610–0.627 for length-based estimates). As a non-destructive approach, the framework preserves specimens for downstream molecular analyses such as DNA barcoding. The modular architecture provides a practical pathway for extending high-precision volumetric estimation through the development of taxon-specific models. Distributed via Docker with a graphical interface and command-line tool, InsectMorphoAI enables reproducible morphometric data acquisition for applications in functional trait ecology, biomass monitoring, and conservation assessments.