An Animal Counting Method Based on Graph Regularized Optical Flow Attention Network
A counting method and attention technology, applied in computing, computer components, instruments, etc., can solve the problems that affect the accuracy of optical flow estimation, not suitable for data sets, etc., and achieve the effect of various animal species
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0037] An embodiment of the present invention provides a method for counting animals based on graph regularized optical flow attention network, see figure 1 , the method includes the following steps:
[0038] 101: Graph regularized flow attention network (GFAN) consists of three parts: shared feature encoder module, count decoder module and temporal consistency module, the network structure diagram is as follows figure 2 shown.
[0039] Among them, the shared feature encoder module uses the VGG-16 network [1]The first four sets of convolutional layers in are used as the backbone to extract the feature map (Feature Map) on two different frames t and t+τ, and then use the online optical flow network based on warping loss to capture the difference between the two frames. sports information. To obtain enhanced feature maps, the generated optical flow is used to warp the feature map of the (t+τ)th frame to the tth frame, where the parameter τ determines the temporal distance be...
Embodiment 2
[0045] The scheme in embodiment 1 is further introduced below in conjunction with specific examples and calculation formulas, see the following description for details:
[0046] 1. Data preparation
[0047] The present invention adopts the video-based large-scale animal count data set AnimalDrone collected by unmanned aerial vehicle in this method during training, and this data set is made up of two subsets, namely AnimalDrone-PartA and AnimalDrone-PartB, and the scene of data set covers different The scene, the variety of animals. After data pruning and annotating the data, AnimalDrone-PartA contains 18,940 images and 2,008,570 annotated objects including training and testing sets, and AnimalDrone-PartB includes 103 video clips including training clips and testing clips, totaling 34704 frames and 2,040,598 annotated objects. Can be used for testing of various target counting methods
[0048] 2. Online optical flow network based on warping loss
[0049] Due to the consider...
Embodiment 3
[0077] The experimental results 1 used in the embodiment of the present invention are shown in Table 1. This result shows the counting evaluation results of various latest methods on the AnimalDrone two-part A and B datasets, including MCNN [2] , MSCNN [3] , CSRNet [4] and other methods as well as the method in the present invention. All counting methods are trained on the training set and evaluated on the test set. The results show that the method used in the present invention can generate more accurate density maps in different situations, and achieve better performance than other methods. Experimental results reflect that the method used in the present invention is superior to existing methods.
[0078]The experimental results 2 used in the embodiment of the present invention are shown in Table 2. The results show the evaluation results of the three variants of GFAN, GFAN-w / o-graph, GFAN-w / o-warp and GFAN-w / o-cnt on the entire AnimalDrone dataset test, in order to bett...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 


