Research On Saliency Detection Algorithm And Its Application In Segmentation Of Images Of Cucumber Diseases

Saliency detection plays an important role in computer vision in recent years.Its purpose is to detect the target area of interest in the image automatically.Detection accuracy and detection efficiency will directly affect the performance of the late target recognition.The focus of this paper is how to improve the accuracy and optimize time complexity of saliency detection algorithm.The saliency detection algorithm proposed in this paper is also applied to the research of cucumber disease processing.The main researches are as follows:(1)In order to solve the problems of edge loss and low detection accuracy in context-aware(CA)algorithm,a new saliency detection algorithm combining prior information and double weights,called PIDWSD is proposed.The algorithm first uses superpixel over segmentation method to divide the image into blocks to get a good target edge,then adds the parameters of Gauss weight and Euclidean distance weight to accurate the saliency map.After combining the center prior information and no saliency association information computing the saliency map,we can remove the interference information in the background.last,we use Sigmoid function to adjust the saliency map,the proposed algorithm has been tested on Berkeley and MSRA1000 image database.The precision-recall curve shows that a higher precision can be obtained than other methods.The experimental result shows that the proposed algorithm effectively solves the edge loss and it's more accurate with detection accuracy 93%in segmentation,it also has a better time complexity.(2)Anew saliency detection algorithm combining manifold ranking and energy equation,called MREESD is proposed to solve the problem of low accuracy and high robustness of the traditional detection algorithm.The algorithm first uses the superpixel segmentation method to divide the image into blocks,puts a new method for calculating the weights of superpixels and selecting saliency seeds to enhance the robustness of algorithm.After combining manifold ranking computing the saliency map to obtain better saliency map,using the energy equation to adjust the saliency map to make it more accurate,also using the threshold segmentation for the obtained saliency map,finally the foreground and background of the original image are separated by adding the binary map to the original image.The proposed algorithm has been tested on MSRA1000 image saliency detection database.The precision-recall curve shows that a higher precision and F-measure can be obtained than other methods.Finally,the experimental results of MREESD and PIDWSD are compared.The results show that the MREESD algorithm has stronger robustness.(3)Segmentation accuracy of crop disease image has a key role in the automatic recognition of disease.In order to solve the problems of low accuracy of the segmentation of cucumber leaf disease image in complex background,we use saliency detection in the processing of cucumber leaf disease images.First,the leaf of cucumber disease was extracted by the saliency detection algorithm.In order to obtain the disease parts,we use ExG to expand the disparity of green parts and lesion parts and then use threshold to the segmentation.Finally,the morphological operation is processed in order to obtain fuller lesion.The proposed algorithm has been tested on common cucumber disease images.The experimental result shows that the algorithm effectively solves the redundant segmentation and it's more accurate with the error rate less than 5 percent in segmentation.On the basis of these,we analysis the characteristics of four typical diseases of cucmber,further extract the characteristics of the disease.Last we use BP neural network classifier to classify the cucumber disease,and achieve recognition rate more than 83%.It verifies the feasibility and practicality of the saliency detection algorithm in processing of disease images.