The Detection Algorithm Research Of Cocoon Appearance Based On Machine Vision

Artificial classification is raw method for cocoon grade evaluation, this method hasaffected the efficiency and accuracy of classification of cocoon, but also greatly damagesthe interests of the silkworms. To solve this problem in this paper, using machine visiontechnology and digital image processing technology to study cocoon appearance detectionand classification algorithm. In this paper, we firstly collect cocoons images and then usedigital image processing algorithm based on OpenCV to filter, binarize, and so on, andcarries on the image segmentation and extraction of characteristic value, and at lastclassification identification is realized by using the BP neural network algorithm. Theexperimental results show that, the cocoon classification has a good effect by usingmachine vision and digital image processing technology.The main results were as follows:(1) The acquisition conditions of Silkworm cocoon image the acquisition conditionsare studied, mainly in light intensity and shooting angle two aspects has carried on theexperiment, using image sharpness evaluation method to evaluate image quality, thecocoon of the image acquisition conditions were determined.(2) The design of cocoon image preprocessing and edge detection algorithm. Amongthem, the pretreatment process includes image filtering and binarization processing. Imagefiltering adopt mean filter, median filter and gaussian filter, three methods of filteringeffects of three kinds of filter method are evaluated, respectively, from a visual and dataanalysis determines the gaussian filter is adopted to cocoon image filtering enhancementprocessing; binarization processing uses the two methods of fixed threshold and automaticthreshold, though comparing and analyzing we find that different cocoon images havingdifferent requirements on the threshold, automatic threshold is more suitable for thecocoon image binarization processing; Edge detection processing adopts three operators ofRoberts operator, Sobel operator and Canny operator, compares the effects of three kindsof operator handling, we determines to deal with the silkworm cocoon image by Cannyedge detection.(3) We choose three kinds of cocoons of different colors and shapes of macularcocoon, yellow cocoon and double cocoon abnormity cocoon as test objects, we analyzethe shape and color cocoons, and extract the shape feature parameters and color featureparameters of silkworm cocoon. In the experiment, we take the operation of the smallestrectangle fitting, etc. for the edge detection image to extract the four shape characteristic parameters of area, duty ratio and the stretching length; we put the original image ofcocoon from the RGB color space into HSV color space to calculate three color featureparameters of the H component, S and V components. We extracting a total of sevenfeature parameters.(4) According to the characteristics, the parameters of hidden layer node number ofBP algorithm is designed7-8-3-3, it realizes the classification of three kinds of differentcolor abnormity cocoon. We Analyze the classification accuracy, improve BP algorithmthat is called conjugate grads BP algorithm, the improved algorithm after three cocoon ofeach type of correct classification rate reached96.8%,96.3%and94.4%, respectively thanbefore improvement increased by19.4%,1.9%and24%, and on the program running timeis saved8seconds time, so we realize the optimization of the algorithm.