[Auto detection on morphological characteristics of termite legs based on deep learning]

The legs of termites follow the standard segmentation pattern of insect appendages. Focusing on the lengths of the femur, tibia, tarsus, and related derived characteristics, we developed a deep learning-based automatic detection and analysis software for termite leg morphology, with a key point detection model for leg segments as the core component to calculate quantitative indicators such as lengths. The results showed that the software could accurately measure the femur, tibia, and tarsus of termite legs, with a mean absolute error of 0.07 mm, a mean relative error of 18.7%, and a root mean square error of 0.09 mm, and could address issues such as occlusion. The positive correlation between hindleg length and body length (Pearson correlation coefficient was 0.413) was higher than that between foreleg length-body length and midleg length-body length, which was consistent with the ecological function of hindlegs. By statistically analyzing the proportional relationships of segment lengths among the foreleg, midleg, and hindleg, we verified the rationality of conventionally using hind tibia length as an indicator for termite leg measurements in morphological studies. This research filled the application gap of deep learning in the field of automatic and accurate measurement of termite leg morphology. Our findings could be extended to studies on other insects with appendage structures.