Research On Cotton Flax Fiber Detection Method Based On Deep Learning

In the textile industry,cotton flax blended yarn is often used in various textile products,and the fiber content ratio of cotton and flax in cotton flax blended yarn will affect the price and quality of cotton flax blended yarn.Generally speaking,the higher the flax content in the blended yarn,the higher the price.In order to prevent the merchants from shoddy,the fiber ratio test is a very necessary test item for the cotton flax blended yarn.At present,the cotton flax fiber content ratio in cotton flax blended yarn is usually calculated manually,which will consume a lot of human resources and time resources.This paper starts with this shortcoming and finds a scheme that can use the network model to automatically detect and count.Firstly,this paper uses fasterrcnn network model and yolov3 network model to train and test cotton and flax fiber data pictures respectively.The results show that fasterrcnn network model is not better than yolov3 network model in accuracy,but yolov3 model is significantly better than fasterrcnn network in detection time.Therefore,this paper selects the yolov3 model as the basic model and optimizes it.The optimization method is to add the attention mechanism module to the network model,and replace and update the learning rate of the network model and the internal activation function.At the same time,this paper notes that fibers are often slender and inclined targets in the image.In order to better detect these fiber targets,the output part of the yolov3 network model is improved,and a new output parameter,angle,is added.Then,the prior bounding box of the network model is changed and the size of the prior bounding box is re clustered to adapt to the aspect ratio and width height value of most fiber targets,After that,this paper adds the angle loss function to the loss function,and adopts the Piou calculation scheme for the IOU calculation of the network output box and the real box.Finally,the cotton and flax fiber detection effect in the picture is visualized and the number of fibers in the picture is counted.The final experimental results show that the optimized yolov3 has a better effect than the original yolov3 model in detecting cotton and flax fibers.The maximum m AP value of the optimized yolov3 model for detecting cotton and flax targets reaches 0.996,and all cotton and flax fibers in a picture can be extracted and counted.However,r-yolov3,which can detect diagonal rectangular targets,does not get better results.The optimized r-yolov3 model can detect cotton The m AP value of flax fiber detection is only 0.788,and there are many omissions in the actual detection.The model still needs to be improved in the follow-up.