Complex environment target segmentation method and device based on multi-module convolutional neural network

Abstract

The invention discloses a complex environment target segmentation method and device based on a multi-module convolutional neural network, and the method comprises the steps: collecting target image data in a complex scene, carrying out the data annotation, and obtaining the annotated target image data; performing data enhancement on the labeled target image data and establishing a data set to obtain a target image data set; performing target preliminary detection processing on target image data in the target image data set based on a convolutional neural network model to obtain a preliminary detection result; and inputting the preliminary detection result into a segmented region shape complementation module, and performing prediction correction on a shielded target through a puzzle algorithm to obtain a segmentation result. According to the embodiment of the invention, the method achieves the precise segmentation of a target in a natural complex environment, and guarantees the segmentation accuracy and real-time performance.

Description

technical field [0001] The present invention relates to the technical field of computer vision, in particular to a method and device for segmenting objects in complex environments based on multi-module convolutional neural networks. Background technique [0002] With the rapid development of computer and automatic control technology, the target instance segmentation technology in the field of computer vision is widely used in all walks of life. It completes pixel-level segmentation on the basis of target detection, which can better meet the needs of daily people's production and life. [0003] However, existing methods usually perform poorly in complex natural environments, mainly due to the following two reasons: 1) Variations in illumination. In natural environments, the intensity and angle of illumination are constantly changing, which leads to changes in the data distribution of images, which affects the corresponding feature extraction. Although this problem can be all...

AI Technical Summary

Problems solved by technology

Although this problem can be alleviated to some extent by data enhancement, it is difficult to effectively perform image correction in an adaptive way

2) Ubiquitous occlusion phenomenon

Target objects are usually occluded by other objects, and incomplete contour information and local texture features pose challenges to existing target segmentation methods, which may cause a complete target to be presented as multiple independent individuals in the visual system, thus being misrecognized For multiple target objects, thereby reducing the accuracy of vision-based target instance segmentation, which severely limits the application of segmentation technology in real scenes

Images