A NOVEL INSTANCE SEGMENTATION ALGORITHM BASED ON IMPROVED DEEP LEARNING ALGORITHM FOR MULTI-OBJECT IMAGES

Abstract

A Deep Learning (DL) algorithm is highly common and attractive in recent years because of its encouraging achievements in many areas. DL lies in image-based detection and instance segmentation of an entity, which is a critical issue that needs further investigation. This paper aims to study the fundamental challenges in using object instance segmentation of images. This paper proposes a novel algorithm for multi-object image instance segmentation algorithm in three stages. A novel backbone approach improves the image recognition algorithm by extracting low and high characteristic levels from the given images in the first stage. The ResNet is the fundamental building block and connects with the Squeeze-and-Excitation Network (SENet) for each ResNet block. The Region Proposal Network (RPN) is used to determine the object item’s placement, followed by the third stage, which suggests an average position RoI layer to choose the optimal boundaries of the instance segmentation. The experiments are conducted and validated using a standard benchmark image dataset, called COCO. The proposed algorithm’s performance is validated using standard evaluation criteria and compared against the recent image segmentation algorithms that use object instances. The results show that the proposed algorithm gets better results than other well-known instance segmentation algorithms in terms of average accuracy over IoU (AP) threshold measures using various thresholds.

Keywords

• deep learning
• multi-object detection
• recognition
• instance segmentation
• average position roi layer