摘要：A pure new MATLAB-based image recognition system was developed to compute the coal particle picture of same grain class through the digital image processing method, 13 image feature parameters was selected to be most representative image characteristic parameters. Take the above parameters as the input of RBF neural network, the density level of coal particles could be estimated ,combined with the real ash content of each density level, the washability curve could be drawed. Experement show,the absolute error of the total ash is 2.375% ,which is Slightly big in the China standards of coal preparation (GB/T477 -1998); the related coefficients of each indicator in both actual and predicted float-and-sink material are all close to 1, while the curves of λ, β, θ and δ are very similar and the deviation of ξ curve is relatively large.