| CPC G06T 7/0004 (2013.01) [G06T 7/11 (2017.01); G06T 7/73 (2017.01); H01M 10/4285 (2013.01); G06T 2207/10116 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20221 (2013.01)] | 12 Claims |
|
1. A battery cell electrode plate detection method, comprising:
performing feature extraction on a to-be-detected battery cell electrode plate image to obtain an electrode plate feature, wherein the electrode plate feature comprises a channel attention feature embedded with position information;
performing electrode plate detection based on the electrode plate feature to obtain an electrode plate detection result, wherein the electrode plate detection result comprises position information of an anode plate and a cathode plate; and
determining an amount of misalignment between the anode plate and the cathode plate based on the position information of the anode plate and the cathode plate in the electrode plate detection result;
wherein the performing feature extraction on a to-be-detected battery cell electrode plate image to obtain an electrode plate feature comprises:
performing a plurality of convolution operations and at least one attention operation on the to-be-detected battery cell electrode plate image to obtain a first feature;
performing multi-scale feature fusion on the first feature to obtain a second feature; and
performing feature fusion from deep to shallow and then from shallow to deep on the second feature to obtain the electrode plate feature;
wherein the performing a plurality of convolution operations and at least one attention operation on the to-be-detected battery cell electrode plate image to obtain a first feature comprises:
performing a slicing operation and then a convolution operation on the to-be-detected battery cell electrode plate image to obtain an initial feature:
performing a plurality of convolution, batch normalization, activation, and cross-stage partial fusion operations on the initial feature to obtain a target partial feature; and
performing an attention operation on the target partial feature to obtain the first feature; and
wherein the performing a plurality of convolution, batch normalization, activation, and cross-stage partial fusion operations on the initial feature to obtain a target partial feature comprises:
performing convolution, batch normalization, and activation processing on the initial feature to obtain a first convolutional feature;
performing a cross-stage partial fusion operation on the first convolutional feature to obtain a first partially fused feature;
performing an attention operation on the first partially fused feature to obtain a first attention-fused feature;
performing convolution, batch normalization, and activation processing on the first attention-fused feature to obtain a second convolutional feature;
performing a cross-stage partial fusion operation on the second convolutional feature to obtain a second partially fused feature;
performing an attention operation on the second partially fused feature to obtain a second attention-fused feature;
performing convolution, batch normalization, and activation processing on the second attention-fused feature to obtain a third convolutional feature; and
performing a cross-stage partial fusion operation on the third convolutional feature to obtain a target partially fused feature;
optionally the performing feature fusion from deep to shallow and then from shallow to deep on the second feature to obtain the electrode plate feature comprises:
performing convolution and upsampling processing on the second feature, followed by feature fusion with the first feature, to obtain an intermediate fused feature;
performing convolution and upsampling processing on the intermediate fused feature, followed by feature fusion with the second attention-fused feature, to generate a first-scale electrode plate feature;
performing feature fusion based on the first-scale electrode plate feature and the intermediate fused feature to generate a second-scale electrode plate feature, and
performing feature fusion based on the second-scale electrode plate feature and the second feature to generate a third-scale electrode plate feature.
|