The Design And Application Of A GPU-based Embedded Fruit And Vegetable Recognition System

The research on the identification technology of fruit and vegetable is of great significance in the field of agricultural production and life.The application of this technology not only promotes the commercialization of the fruit and vegetable industry,but also reduces the labor cost.Moreover,the artificial intelligence recognition technology is the identification of other artificial intelligence.And detection techniques provide a theoretical and practical reference value.Fruit and vegetable identification technology belongs to image recognition technology.At present,the main methods of image recognition include a traditional pattern recognition method and a deep learning pattern recognition method.The traditional image recognition method has the disadvantage that the recognition accuracy is not very high,and it is necessary to manually describe the characteristics of the detection target to the machine,and the detection model has poor anti-interference ability and high equipment cost,which is difficult to meet the practical application requirements of people.However,deep learning has solved the above problems well.In this paper,the deep learning technology is applied to the fruit and vegetable identification technology.Firstly,the deep convolutional neural network method is used to realize the detection task of fruit and vegetables on the PC.Then the algorithm is transplanted to the embedded development platform Jetson TX2 to realize the fruit.Collection,processing and detection of vegetable image data.The fruit and vegetable identification system of this paper effectively solves the problem that the detection algorithm is difficult to identify and locate the fruits and vegetables due to the complex environment of fruit and vegetable growth,and different light intensity factors,etc.,for tomato detection,peach detection and classroom personnel detection.,proposed a multi-stage detection network.Considering the detection performance and speed,comparing the different labeling methods,different network input sizes,different network structures and different test results of the training sample IOU standard,a multi-stage detection network is preferred.The results showed that the detection accuracy of the detector for mature tomato was 89.3%,the detection speed was 0.25s/image,the detection accuracy of peach was 92.3%,and the detection speed was 0.31s/image.Then the accuracy of the test algorithm is extended to 98.5%,and the recall rate is 91.2%.