Existing methods for detecting road damage mainly depend on manual inspections or sensor-equipped vehicles, which are inefficient, have limited coverage, and are susceptible to errors and delays. These traditional methods also struggle with detecting minor damage, such as small cracks and initial potholes, making real-time road monitoring challenging. To address these issues and improve the performance for real-time road damage detection using Street View Image Data (SVRDD), this study propose DAPONet, a new deep learning model. DAPONet proposes three main innovations: (1) a dual attention mechanism that combines global context and local attention, (2) a multi-scale partial overparameterization module (CPDA), and (3) an efficient downsampling module (MCD). Experimental results on the SVRDD public dataset show that DAPONet reaches a mAP50 of 70.1%, surpassing YOLOv10n (an optimized version of YOLO) by 10.4%, while reducing the model’s size to 1.6 M parameters and cutting FLOPs to 1.7 G, resulting in a 41% and 80% decrease, respectively. Furthermore, the model’s mAP50-95 of 33.4% on the MS COCO2017 dataset demonstrates its superior performance, with a 0.8% improvement over EfficientDet-D1, while reducing parameters and FLOPs by 74%.

Medical image classification systems are increasingly expected to provide accurate predictions while remaining computationally practical. However, conventional deep learning models typically process all samples through the same inference path, regardless of whether a case is straightforward or ambiguous. We propose Computer-Aided Triage and Diagnostic Routing (CATDR), a conditional inference framework that combines lightweight screening with selective downstream expert escalation. An EfficientNet-B0 screening model first evaluates each image using a softmax-derived confidence score. Samples exceeding a predefined screening threshold are finalized at the screening stage, whereas the remaining samples are forwarded to a learned routing module that selects a downstream expert model. Auxiliary output-head selection is retained as a logical prediction-selection mechanism rather than a physical early-exit mechanism. CATDR was retrospectively evaluated on three public medical imaging datasets covering fetal ultrasound, gastrointestinal endoscopy, and dermoscopic skin lesion classification. Under the primary evaluation protocol, CATDR achieved accuracies of 94.40%, 93.10%, and 73.80% on FPDB, Kvasir v2, and PAD-UFES-20, respectively, with corresponding mean inference times of 4.60, 4.92, and 14.51 ms. Calibration analysis further showed that development-set temperature scaling reduced the expected calibration error from 0.0420 to 0.0128 on FPDB, from 0.0376 to 0.0192 on Kvasir v2, and from 0.1578 to 0.0368 on PAD-UFES-20. These results support the technical feasibility of confidence-guided screening and selective expert escalation for medical image classification while highlighting the importance of calibration and operating-point selection.