<p>Accurate skeletal age estimation is essential in medicolegal investigations, particularly when identification records are unavailable or suspected of tampering. Unlike other joints, the pelvis remains informative throughout adolescence and early adulthood, exhibiting strong sexual dimorphism. However, traditional age-estimation techniques based on the pelvis rely heavily on experts and are highly subjective, prone to inter-observer differences. Furthermore, low contrast, high inter-variability between foreground and background, variability in sex-specific age patterns, and anatomical overlap in pelvis X-rays make it difficult to extract automatic features. To overcome these limitations, this paper presents PelvisNet, a new sex-specific deep learning framework for estimating skeletal age from pelvis radiographs. The proposed framework consists of Contrast Limited Adaptive Histogram Equalization (CLAHE) for low-contrast image enhancement, Region-Based Symbolic Segmentation (RSSeg) for accurate extraction of high inter-variability in the foreground and background of the pelvis region, and an EfficientNetB5-based network for hierarchical feature learning, effectively extracting anatomical overlap features. Additionally, the sex-specific age pattern aids sex classification, and probabilistic information is then combined with deep features to estimate age groups. The framework was evaluated on a database of 1,837 pelvis radiographs obtained from primary and secondary sources across age categories relevant to medicolegal practice. The accuracy of sex-based age group estimation reached 91.67%. The proposed framework offers a reliable, clinically interpretable decision-support tool for medicolegal age estimation, where accuracy and reliability are important, particularly across overlapping adolescent age groups.</p>