Research On Target Detection Of Sterile Rice Ears Based On Deep Learning

For half a century,the hybrid advantage of rice has been fully utilized for the benefit of the world.However,as our country has progressed in technological research on rice,it has been slow in developing new sterile rice.With the need to continuously breed new sterile rice varieties and the declining number of people working in agriculture,it has become increasingly difficult to find and check for new sterile rice.To help rice researchers accurately detect sterile rice in rice fields,this paper designs a target detection system for sterile rice ears based on the open-source framework Pytorch,which obtains multiple information about sterile rice ears by user uploading photos to improve detection accuracy and thus reduce the intensity of manual identification.The main accomplishments of the paper are as follows.(1)To address the problem of scarcity of sterile rice ears data set,we use the pictures of sterile rice ears and fertile rice ears collected in rice fields as the original data set,and then obtain enough rice ears pictures from the Internet,and use image synthesis technology to synthesize some of the actual pictures taken with these pictures obtained online as part of the original data set.The original dataset was later increased to 8000 images by data expansion to meet the research requirements.(2)For the problem of rice spike detection,the accuracy requirements of the sterile rice spike target detection task can be met by conducting control tests based on three sets of rice spike target detection models: Faster RCNN,SSD,and Retina Net.The detection accuracies of sterile rice ears were 98.31%,95.43%,and 95.97%,respectively.The Faster RCNN model with the highest detection accuracy was selected,and then by selecting different feature algorithms and optimization algorithms,the m AP value of the Faster RCNN-VGG16-Adam model was obtained as 96.75%,and the FPS in terms of real-time performance was 43.1.(3)The design and implementation of the sterile rice spike detection system were carried out to address the inconvenience of sterile rice spike detection.Based on the SSH framework,the sterile rice cob detection system was designed and implemented to provide a front-end interactive page for users to realize sterile rice detection and location indication functions.