Inspection of combs is a critical task in beekeeping for assessing comb condition and bee population size. Manual counting is labor-intensive and error-prone, highlighting the need for automation. We propose a machine-learning-based method using single comb images that requires no specialized equipment. We annotated three honeybee parts -- abdomen, head, and whole-body -- and fine-tuned SSD and YOLOv11 models to detect each part. Predicted bounding boxes were integrated using the Hungarian method to compensate for occlusion-induced missed detections. The abdomen detector achieved superior accuracy across all density levels, and the integrated approach effectively reduced false negatives. Applying the matching framework to YOLOv11 confirmed its detector-agnostic nature, demonstrating consistent recall improvements over whole-body detection alone.