Feature Extraction Of Complex Texture Fabric Opening And System Development

In view of the current situation that the efficiency of fabric manual opening process is low,domestic equipment has not yet stable products to enter the market,and foreign semi-automatic equipment operation is relatively complex and general,this paper focuses on improving the universality of automatic opening equipment with the title of "feature extraction and system development and application of fabric opening with complex texture background".Combined with deep learning algorithm and traditional image processing method,the automatic opening of fabrics with different texture characteristics is completed.The main algorithm design of the automatic opening system studied in this paper is divided into two parts: one is to use deep learning to complete the analysis and classification of fabric texture characteristics;the other is to use targeted image processing algorithm to complete the accurate recognition of fabric opening guide wire according to the classification results of fabric texture characteristics.The main research contents and achievements of this paper are as follows(1)According to the analysis of the existing algorithms of the automatic opening system,the texture characteristics of the fabric itself will affect the recognition of the opening line due to the complexity of the fabric itself.Therefore,based on the existing 1500 fabric image samples,this paper constructs a fabric texture feature data set,which is divided into seven different texture features,such as dense pattern,equal width transverse pattern,unequal width transverse pattern and so on.Using the improved googlenet network learning technology based on transfer learning,seven texture features are classified,and the final classification accuracy reaches 100%.Based on the trained network,the analysis and classification system of fabric texture characteristics is constructed,and 300 fabric image samples outside the data set are introduced to verify the generalization ability of the network.The final recognition accuracy reaches 98%,which shows the reliability of the classification system and the feasibility of the classification algorithm.(2)After finishing the classification of fabric texture characteristics,according to the classification results,the traditional image processing algorithm is improved to complete the recognition of fabric opening guide line.For example,the dense pattern with simple texture characteristics can be recognized by using vertical projection and Otsu segmentation algorithm,and the final recognition accuracy reaches 100%.However,the lattice pattern and stripe lattice with relatively complex texture characteristics can not be recognized accurately by the dense pattern recognition algorithm.In this paper,the improved method is adopted for the fabric The texture filtering algorithm combined with the color clustering algorithm based on color difference is used to recognize the fabric opening guide line under the texture characteristics,and the final recognition accuracy reaches 98%.(3)There are three types of open width guide lines: single stitch,double stitch and missing stitch.The features of single stitch and double stitch are relatively simple and fixed,while the features of missing stitch belt are relatively complex and not fixed.(2)the algorithm in this paper is based on the recognition of single stitch and double stitch of fabric opening guide line.If the fabric opening guide line is missing stitch belt,an active contour model should be added on this basis.First,an initial curve is given near the region of interest,and the curve is expressed As an energy functional function,the energy functional is minimized to make the curve approach the target contour continuously in the image under the internal force(control curve bending and stretching)and external force(image brightness and gradient),and finally converge to the edge contour of the target image.The final recognition accuracy of the missing needle tape is 98%.(4)According to the requirements of field production speed: 120 m / min,opening accuracy: ± 3mm,complete the selection of hardware required for the construction of the system,the selected camera is Germany Yingmei precision industrial camera DFK z12gx236,the selection of motion control part is: Panasonic PLC fp-xh c60 t and delta asda-b2 series servo driver.The test of the system is divided into laboratory hardware and software test and field test.The laboratory test results show that the selection results can meet the production needs.The field test results show that the maximum opening average deviation of all kinds of opening guide lines is 1.0 mm,the maximum offset error is less than 1.5 mm,and the frequency is very low,which can meet the preset accuracy requirements.The maximum average recognition time of the system is 86 MS,which can meet the speed requirements of the system.