Optimization Of Identification Software System For The Liquid Drop Fingerprint

The liquid drop fingerprints are curve images obtained after filtering and normalizing,representing liquid characteristics by collecting the optical fiber capacitance data in the process of droplet formation by the fiber-capacitive drop analyzing technology.By identifying the liquid drop fingerprint,the type of liquid can be identified and the characteristic of liquid can be analyzed.Liquid drop fingerprint recognition software system mainly consists of three parts:signal preprocessing module,feature extraction module and pattern recognition module.In this paper,with the theory and the experimental data analyzed,these three modules are optimized and the final recognition results are verified by experiments.In the fiber-capacitance droplet analysis system,some uncertain factors lead to the difference between measured value of the sensor capacitance and the actual value.In order to effectively solve the problem,this paper describes a drift correction algorithm.Theoretical analysis and experimental results show that after modified by the correction algorithm,capacitance measurement value reflects the nature of the droplet itself and is comparable.The initial capacitance value variance reduces from 4.8×10-2 to 1.4×10-4,improving the stability of the system and providing valid data for future study.In order to effectively detect the abnormal liquid drop of the droplet analysis system and improve the accuracy of the liquid drop fingerprint,a new anomaly detection method based on boxplot is put forward.Experimental results show that the detection recognition ratio of the abnormal liquid drop can be ensured after feature optimization,together with the greatly reduced computational complexity.Boxplot method is effective in detection of abnormal liquid drop fingerprint,with its accuracy up to 100%among selected samples.In order to effectively reduce the time complexity of clustering algorithm,a new method based on multiple linear regression is put forward to reduce the eigenvector dimensions of the liquid drop fingerprint.Experimental results show that the recognition ratio of the liquid drop fingerprint can be ensured,together with the reduced computational complexity and excellent clustering accuracy.Compared with hierarchical clustering method,the iterative dynamic clustering method is more effective in liquid identification,with its accuracy up to 100%among selected samples.In order to effectively reduce the time complexity of the recognition algorithm and improve the recognition accuracy and the generalization ability,a new method combining support vector machine with clustering is put forward.Experimental results show that the recognition accuracy can be up to 100%among selected samples,together with the reduced computational complexity of training models and the significantly improved recognition efficiency.Compared with the previous model,the generalization capability has been greatly enhanced,with its estimation of generalization performance been cut exceeded 90%.