In order to improve the accuracy of water surface garbage identification

a modified convolutional neural network model of Caffe Net is proposed to identify water surface garbage. The model improves the size of the convolution kernel

the number of convolution kernels

and adds a sparse structure

which enhances the ability of network model feature extraction and reduces network complexity. The experimental results show that the improved Caffe Net model improves the recognition rate of water surface garbage to 95.75 %

which can reduce the influence of water surface ripple

object reflection and bridges on the recognition of water surface garbage

and has better water surface garbage recognition effect.