Salient object detection method in complex environment

Abstract

The invention discloses a salient object detection method in a complex environment, and the method comprises the steps: constructing a convolutional neural network at a training stage, wherein the convolutional neural network comprises an input layer, a coding part, a decoding part and an output layer, the input layer comprises an RGB image input layer and a thermodynamic diagram input layer, the coding part comprises 10 neural network blocks, the decoding part comprises five cross modal fusion blocks, five bilateral inversion fusion blocks and one multi-scale consistency fusion block; inputting the RGB images of each pair of original 3D images in the training set and the corresponding thermal images into a convolutional neural network for training, and outputting saliency prediction images corresponding to each pair of original 3D images; obtaining an optimal weight vector and an optimal bias term of the convolutional neural network training model through multiple iterations and calculation of a loss function value between the saliency prediction image and the corresponding label image; and performing predicting by using the optimal weight vector and the optimal bias term during testing to obtain a saliency prediction image; the method has the advantage of high significance detection precision.

Description

technical field [0001] The invention relates to a visual image saliency detection method based on deep learning, in particular to a saliency object detection method in a complex environment. Background technique [0002] With the rapid development of deep learning in the computer field, image saliency detection has become a research field that has attracted more and more attention. Salient Object Detection (SOD) aims to distinguish the most visually distinctive objects from the input image, and it is an important feature of many image processing and computer vision tasks (such as face recognition, video compression, image editing, semantic segmentation, etc.) etc.) an efficient preprocessing step. Traditional salient object detection methods perform poorly and are limited by hand-crafted relevant features, and salient object detection has been greatly developed with the rise of convolutional neural networks. The salient object detection methods in the past few years all us...

AI Technical Summary

Problems solved by technology

The existing two-stream salient object detection method using RGB image and thermal image also adopts the encoding-decoding structure. The multi-level features extracted from the two modalities are combined for salient object detection, but the detection performance of such two-stream salient object detection methods needs to be further improved

Images