Ice Scoop Block Texture Segmentation Method Based on Deeplabv3+

In response to the issues of poor robustness and low detection efficiency of conventional ice scoop defect detection algorithms, this paper proposes an improved Deeplabv3+ ice scoop defect detection algorithm, which is used for the detection of black knots and pollution blocks on ice scoops. The new model replaces the feature extraction network with Mobilenetv2 to decrease the parameters and adds Bisnetv1 feature fusion module to enhance the combination of shallow feature and deep feature. To verify the effectiveness of the algorithm, experiments were conducted using a self-made dataset. The experimental results showed that the MIoU (Mean Intersection over Union) of the improved Deeplabv3+ on the test set reaches 80.89% and the parameters is only 9.8% of original model, making it more suitable for use in ice scoop detection systems.