Tungsten concentrate flotation image working condition identification method

A tungsten concentrate flotation froth image working condition recognition method comprises the steps that a froth data set is divided into five working condition states, after image preprocessing and image enhancement are conducted, a tungsten concentrate froth image data set is trained through an improved resnet50 network, and according to the degree of accuracy of a training model, the working condition of a froth image is recognized. And selecting the working condition recognition model with the highest accuracy as an offline model, and performing recognition and classification on the input foam image. In an improved renet50 structure, a convolution block is used for changing the number of feature map channels, and a channel attention module is added to a residual block after convolution is completed to achieve deep extraction of foam features. The method enables the training model to have a high recognition rate for the foam image, improves the discrimination capability for the working condition state, a