Camouflaged Insect Segmentation Using a Progressive Refinement Network

Accurately segmenting an insect from its original ecological image is the core technology restricting accuracy and efficiency of automatic recognition. However, performance existing segmentation methods unsatisfactory in images shot wild backgrounds on account challenges: various sizes, similar colors or textures to surroundings, transparent body parts vague outlines. These challenges are accentuated when dealing with camouflaged insects. Here, we developed method based deep learning termed progressive refinement network (PRNet), especially for Unlike methods, PRNet captures possible scale location insects by extracting contextual information image, fuses comprehensive features suppress distractors, thereby clearly Experimental results 1900 demonstrated that could effectively segment achieved superior detection performance, a mean absolute error 3.2%, pixel-matching degree 89.7%, structural similarity 83.6%, precision recall 72%, which improvements 8.1%, 25.9%, 19.5%, 35.8%, respectively, compared recent salient object methods. As foundational detection, provides new opportunities understanding camouflage, also has potential lead step progress intelligent identification general insects, even being ultimate detector.