Research On Segmentation And Recognition Of Forage Image Based On Color Enhancement

In livestock,agriculture and grassland management projects,grassland forage identification is one of the most important aspects of grassland digitisation.At present,forage identification mainly relies on manual identification,which requires not only expert resources but also field collection of samples,which is time-consuming and inefficient.Therefore,this thesis takes forage as the research object,collects forage images in the field and establishes a database to study a fast segmentation recognition and classification method for forage images.The main research content is as follows.(1)Collect and establish forage image data set.The forage image data sets in this thesis are taken in natural light by Canon camera EOS60 d.A total of six types of forage images were collected,including Agropyron Cristatum(L.)Gaertn,Medicago Polymorpha L,Elymus Sibiricus Linn,Bromus Inermis Leyss,Mengnong Shamrock and Onobrychis Viciifolia Scop.Considering the difference in time complexity and spatial complexity of the algorithm,in addition to creating a dataset related to forage in a simple context,a dataset of forage images in a complex context of grassland was created in contrast to it.(2)A simple background forage image segmentation and recognition method based on color enhancement is proposed.Firstly,the color space of the original forage image is converted,and a new color palette is constructed according to the colors accounting for the top 85% in the image color frequency histogram.The distance from the remaining pixels in the image to the color in the palette is calculated,and the nearest color is selected to cover it to generate a feature map.Hierarchical clustering and K-means clustering are used for feature image pixel points.Taking the Hu invariant moment of the outer contour of the clustered forage image as the eigenvalue,through the random forest decision recognition,and adding the 6 fold cross validation,the experimental results show that the recognition accuracy is improved by 18%-46%.(3)A forage image segmentation method with complex background based on LUV-L component and Grab-Cut is proposed.Firstly,preprocessing operations such as denoising and contrast enhancement are carried out on the forage image,then the feature map of L channel is extracted after the forage image is converted to LUV color space.Finally,the processed feature map is segmented by Grab-Cut method.The average ratio of the target area segmented from the experiment to the target area segmented manually is 91.64%.Compared with FCM,Extreme C3,Deeplabv3 + and other methods,the segmentation effect is better.