Prediction Of Colored Solution Concentration Based On Machine Vision And The IDE-BPNN

Solution concentration is a common and important parameter in the process of industrial production.In order to improve the quality of product and reduce the costs of production,it often needs to be real-time detected.Most of available online sensors should be placed under the solution to finish the measurement of concentration,but for some corrosive or other liquid which is not suitable to contact directly,this kind of sensor has certain limitation.Considering that a lot of solution is colored solution,and the color can indirectly reflect concentration,a kind of non-contact concentration measurement method for the colored solution based on machine vision was put forward.The method firstly obtained colored images of solution by using the image acquisition equipment,then based on the BP Neural Network(BPNN),the mathematical model which can reflect the function relationship between three color values of the colored image and the concentration of colored solution was built,and using the Improved Differential Evolution(IDE)algorithm to calculate the connection weight and threshold of neural network.The concrete research content is as follows:First of all,an improved differential evolution algorithm was put forward,it mainly included improvements in mutation strategy and the selection strategy,and its performance was tested based on standard test functions.Test results showed that the IDE algorithm was far better than that of DE algorithm in terms of precision,convergence speed and stability.Then a method to build concentration prediction model of the colored solution based on IDE-BPNN was presented,using the BPNN to build the concentration prediction model and using the IDE algorithm in neural network training.Finally,the potassium permanganate solution was used to make experimental verification to test above method.The images of colored solution which was configured according to the concentration were obtained through image acquisition equipment,and the images were processed,including removing the noise of images and reading three color values.The three color values were used for training and testing the IDE-BPNN algorithm to verify the feasibility of IDE-BPNN algorithm,the results showed that,the error of the method which was based on machine vision and the IDE-BPNN algorithm to predict concentration of colored solution was very small,and it proved the feasibility of this method.In order to highlight the advantage of IDE-BPNN algorithm,based on DE-BPNN algorithm and BP-BPNN algorithm,concentration prediction model was constructed and the same solution concentration was predicted,concentration prediction result was compared with the above method,comparison result showed that the concentration prediction result based on the IDE-BPNN algorithm was superior to other two methods.