Research On Seed Recognition And Classification Of Jatropha Curcas Based On Machine Vision

In order to solve the problems of traditional artificial methods for screening Jatropha seeds,such as strong subjectivity,high error rate and low efficiency,in this study,in the laboratory environment,according to the requirements of machine vision system design,the existing equipment was used to identify the Jatropha seeds.In the experimental process,targeted research work was conducted on the problem of connected domain noise in the image processing part and the problem of excessive data redundancy in the feature parameter extraction and processing part,and the comparative application of the algorithm was made locally.At the same time,the robustness of the machine vision system designed in this experiment is improved overall.In the experimental process,the collected seeds of Jatropha,using Open CV software and Matlab software to perform image preprocessing and feature parameter extraction on the image respectively,combined with the support vector machine classifier of particle swarm optimization algorithm to achieve high efficiency,fast,accurate identification and classification of Jatropha seeds.The main research content of the thesis has the following parts:1.Image processing:when filtering the Jatropha seeds image,there is connected domain noise,which will have a certain impact on the extraction of feature parameters,in this paper,based on the Otsu algorithm,the threshold segmentation algorithm of the Fill Internal Contours is applied,that is,the combination of the contour coverage based on the internal small connected domain and the threshold segmentation method,according to the distribution of different grayscale values in the image,set a threshold,divide it into two grayscale areas,and divide the pixel areas of different grayscales,binarize intotwo areas.The area threshold is set on the basis of the binarized image.The Fill Internal Contours algorithm is used to fill in the connected domain less than or equal to this threshold,which can effectively eliminate the interference of noise.2.Feature extraction: in order to fully obtain the information of Jatropha seeds image,this paper extracts and analyzes the shape feature,color feature and texture feature of Jatropha seeds image.Shape characteristics include area,perimeter,long axis,short axis,equivalent diameter,roundness,elongation,compactness,rectangularity,eccentricity and other characteristic parameters;the color characteristics are mainly(Red)R,(Green)G,(Blue)B,(Hue)H,(Saturation)S and(Value)V components;the texture features are mainly parameters such as mean,contrast and entropy.According to the corresponding feature parameter extraction algorithm,a total of 19 feature parameters were extracted and a data comparison database was established,which laid the foundation for the feature identification and classification of Jatropha seeds.3.Classifier selection: By principal component analysis(PCA),select 5 out of 19 feature parameters of the Jatropha seeds image(such as long axis length,compactness,B component,H component and mean)can represent most of the image information,This can shorten the processing time of the classifier.Through the support vector machine(SVM)of Particle Swarm Optimization(PSO),that is,to find the optimal parameters of the SVM model,construct the PSO-SVM model,the accuracy rate of the identification and classification of the seeds of Jatropha reached about 96%.The experimental results show that using the threshold segmentation algorithm based on Otsu and Fill Internal Contours under the same conditions can improve the processing effect on the noise of large-area connected domains than the traditional method,and avoid the blindness of morphological processing on such noise processing capabilities.At the same time,after adding the particle swarmoptimization algorithm to optimize the support vector machine parameters,the effect of data classification and further identification is correspondingly enhanced,which proves that the support vector machine classifier of the particle swarm optimization algorithm is effective in this experiment.