Deep Learning-based Recognition And Counting Of Wheat Ears At Maturity

Wheat is an imperative food crop,and its yield directly affects the global food supply and price.Therefore,achieving accurate wheat yield estimation is of great significance in food security,economic development and agricultural production management.The number of wheat ears is one of the factors of wheat yield estimation,and the ripening period is the last stage of the wheat growth cycle,and obtaining the number of wheat ears accurately in the ripening period is of great significance for wheat yield estimation.With the application of deep learning technology in the field of agriculture,the research on the use of object detection algorithm for wheat ears recognition and counting has received growing attention.In this experiment,a research on the recognition and counting of mature wheat ears based on deep learning was carried out,in order to realize the rapid and effective recognition and counting of wheat ears in the field and promote the development of agricultural informatization.The main work of this research is as follows:(1)Build a wheat dataset with multiple data sources.In addition to conventional field data collection and public data,the experiment supplemented the collection of data from simulated wheat field scenes using materials such as wheat plants and flower mud boards,solving the problems of location limitations,season limitations,and adverse weather effects on field data collection.By enhancing the data,a high-quality wheat dataset was obtained.(2)Construction of wheat recognition counting network.To construct an accurate and reliable wheat recognition counting network,five models,Faster R-CNN,Retina Net,SSD,YOLOv5,and YOLOv7,were used in the experiments,and the datasets composed of multichannel data were input into the models for training,and the five models were evaluated quantitatively at the end,and the wheat recognition counting was further evaluated by the fixed area of wheat images collected in the field The accuracy of the network was verified,and the results showed that the accuracy of all five models exceeded 90%,among which the accuracy of YOLOv7 reached 97.04%,and by optimizing and improving two models,YOLOv5 and Faster R-CNN,the accuracy of the two models improved by 1.46% and 1.49% to 97.3% and97.14%,respectively,further improving the wheat accuracy of the recognition network.(3)Design and development of an integrated management system for wheat crop identification and counting.In order to assist wheat yield estimation,promote the wide application of smart agriculture technology,and make the results of smart agriculture really popular to the users,this paper relies on the experimental wheat ears identification and counting network,and combines the needs in the process of wheat cultivation and planting production,and designs and develops a comprehensive management system of wheat ears identification and counting with the functions of wheat ears identification and counting and mu ears calculation.