US 12,169,958 B1
Method, system and device for acquiring seaweed bed area, and storage medium
Fan Li, Yantai (CN); Shaowen Li, Yantai (CN); Zhaowei Liu, Yantai (CN); Haixia Su, Yantai (CN); Huawei Qin, Yantai (CN); and Zhanyu Wang, Yantai (CN)
Assigned to Shandong Marine Resource and Environment Research Institute, Yantai (CN); and Yantai University, Yantai (CN)
Filed by Shandong Marine Resource and Environment Research Institute, Yantai (CN); and Yantai University, Yantai (CN)
Filed on Apr. 23, 2024, as Appl. No. 18/643,003.
Claims priority of application No. 202410121538.1 (CN), filed on Jan. 30, 2024.
Int. Cl. G06V 10/50 (2022.01); G06V 10/60 (2022.01); G06V 20/10 (2022.01)
CPC G06V 10/50 (2022.01) [G06V 10/60 (2022.01); G06V 20/188 (2022.01)] 6 Claims
 
1. A method for acquiring a seaweed bed area, comprising the operations of:
S1, acquiring an initial seaweed bed image of a region to be measured, and removing a fog thickness in the initial seaweed bed image to obtain a fog thickness removal chart; performing a first quality evaluation on the fog thickness removal chart based on a brightness to obtain a first score, and judging whether the first score exceeds a first score threshold; if so, executing S3; if not, executing S2;
the operation of removing the fog thickness in the initial seaweed bed image comprises: obtaining a transmittance corresponding to each position point based on an acquired fog thickness of each position point in the initial seaweed bed image; and subtracting a first atmospheric light value from an initial pixel value of each position point in the seaweed bed image, and dividing a resulting value by the transmittance of the corresponding position point to obtain the fog thickness removal chart;
wherein the operation of obtaining the first score comprises dividing the fog thickness removal chart into a plurality of block regions, and performing multi-filtering on all the block regions to obtain a filtered region block set; obtaining a brightness similarity between each filtered region block and a corresponding initial block based on a brightness mean value of each filtered region block in the filtered region block set and a brightness mean value of a corresponding initial block of the each filtered region block in the initial seaweed bed image; respectively multiplying all the brightness similarities by a weight value of a corresponding filtered region block and then summing to obtain a brightness quality score which serves as the first score;
S2, removing a fog concentration in the initial seaweed bed image to obtain a fog concentration removal chart; performing second quality scoring processing on the fog concentration removal chart based on a pixel standard deviation to obtain a second score, and judging whether the second score exceeds a second score threshold; if so, executing S3; if not, executing the S2;
wherein the operation of removing the fog concentration in the initial seaweed bed image comprises: obtaining a diffraction rate of each position point based on an acquired minimum value of a dark channel of the initial seaweed bed image; obtaining the diffraction rate of each position point by an equation: Tj=1−wj·a, where Tj is the diffraction rate of position point j, wj is a weight value of the position point j, and a is a minimum value of the dark channel; obtaining the fog concentration of each position point based on the diffraction rate and a pixel value of each position point; obtaining the fog concentration corresponding to each position point by an equation: Aj=p0j/Tj+b, where Aj is a fog concentration of position point j, p0j is an initial pixel value of position point j, and b is a second atmospheric light value; subtracting the pixel value of each position point from the fog concentration in the initial seaweed bed image, and dividing a resulting value by the diffraction rate of the corresponding position point to obtain the fog concentration removal chart;
the operation of obtaining the second score comprises: dividing the fog concentration removal chart into a plurality of block regions, and performing multi-filtering on all the block regions to obtain a fog concentration removal filtered region block set; obtaining a contrast similarity between each fog concentration removal filtered region block and a corresponding initial block based on the pixel standard deviation of each fog concentration removal filtered region block in the fog concentration removal filtered region block set and a pixel standard deviation of the corresponding initial block of each fog concentration removal filtered region block in the initial seaweed bed image; and respectively multiplying all the contrast similarities by a weight value of the corresponding fog-concentration-removal filtered region block, and then summing to obtain a fog concentration removal contrast quality score as the second score;
S3, acquiring low-level semantic features and high-level semantic features of the fog thickness removal chart or the fog concentration removal chart, and after fusion, performing receptive field expansion processing to obtain a first semantic feature graph; after depth scaling the first semantic feature graph, performing global feature extraction to obtain a second semantic feature graph; fusing different scales of pooling graphs of the second semantic feature graph to obtain the feature seaweed bed graph;
S4, acquiring a pixel change value of each position point compared with a previous position point in the feature seaweed bed graph, obtaining a pixel difference corresponding to each position point, all the pixel differences and all the corresponding position points, and forming a pixel difference distribution map; acquiring, as an edge point, a position point in the pixel difference distribution map where the pixel difference is greater than a pixel difference threshold; obtaining a seaweed bed region based on all the edge points; and
S5, obtaining a seaweed bed area based on the seaweed bed region and an image scale.