Apple Target Recognition Algorithm Based On Saliency Detection And Deepsnake

Apple is the main agricultural cash crop in China.In recent years,the agricultural labor force has been rapidly transferred to other industries.In the busy season of farming,the majority of rural areas have begun to face labor shortage,and the production efficiency has been significantly reduced.The contradiction between the rapid development of fruit and vegetable production and the shortage of agricultural labor force and excessive labor intensity has become increasingly apparent.Apple-picking is a huge job,and robots are needed to further improve efficiency and quality.Therefore,the research and application of fruit crop picking robot is of great significance to reduce the labor intensity of agricultural practitioners,liberate agricultural labor force and improve the intensive production level of fruits and vegetables.The primary task of the picking robot is to identify and locate the ripe apple target by using the visual system.Based on the relevant algorithms in the field of image processing research,to realize the automatic detection,recognition and positioning of fruits,has become the current research hotspot in the development and application of automatic fruit picking robot.In this context,it is particularly important to carry out optimization research on fruit recognition and detection algorithm.The fruit recognition and detection algorithm not only provides theoretical and technical means for the application of the fruit recognition algorithm in the apple picking robot,but also to some extent makes up for the high cost,high time consumption and other deficiencies,while ensuring real-time and improving the recognition accuracy,to ensure the efficient work of the apple picking robot.The main research contents of this paper are as follows:1.In order to solve the existing technology of high algorithm complexity and(or)operation for a long time,has not been able to better achieve real-time adaptive picking operation problem,put forward a kind of fruit used in apple picking robot target recognition algorithm to realize low cost,high accuracy,good real-time performance,improve the work efficiency of apple picking robot.Based on global luminance difference calculation and color aberration method,the salience of the image was obtained.Adaptive image segmentation and maximum connected domain analysis were used to detect the target fruit region.On this basis,the histogram backprojection algorithm is used to further optimize the salient graph results combined with the original image.2.Targeting the overlapping fruit areas that may exist in the salient map.A dynamic adaptive overlapping target separation algorithm is proposed,and the single target fruit is located by maximal connected domain analysis.Finally,the initial contour obtained in the above steps is used to obtain the final target fruit through real-time Deep Snake instance segmentation.The method can realize the real-time detection,positioning and recognition of fruit targets,which provides the prerequisite for the implementation of the mechanical action of the picking robot,and realizes the effective separation of overlapping fruits,which greatly improves the working efficiency of the apple picking robot.A self-built data set containing 1036 images of Apple in natural environment was tested,the FMeasure value calculated by the method in this paper based on the segmentation results was 91.90%.The mean IOU calculated from the test results was 0.85 and the standard deviation was 0.14.For non-overlapping shading fruits,overlapping fruits,shading branches and leaves,and poor light conditions,the detection accuracy of target fruits were 99.12%,94.78%,90.71%,94.46%,respectively.The comprehensive detection accuracy was 95.66% In the 1036 test images,the average processing time was 0.42 s.3.1036 images from the natural environment were collected in Yiyuan County,Zibo City,Shandong Province to form a test data set and 300 training data sets.The self-built 1036 pictures of apples in natural environment include apples of various varieties,apples with graphic labels and plastic bags,pictures collected under different lighting conditions,and the pictures include single unshaded apples,multiple apple fruits,as well as the cases with branches and leaves shaded and fruits overlapping.