Research On Cotton Yield Prediction System Based On Improved YOLOv5

Cotton is an important cash crop in China,which has important applications in agricultural and industrial production.Due to its vast geographical area and unique geographical advantages,Xinjiang region ranks first in cotton cultivation in China,accounting for 83.22% and 90.20% of the total planting area and yield in the country.The prediction of cotton yield helps in economic regulation and adjustment of planting patterns,improving production returns,while traditional manual yield measurement methods require a significant amount of time and labor costs.To solve this problem,this article selects cotton images after spraying defoliants as the research object,constructs relevant datasets,uses deep learning neural networks and cotton yield mathematical models as the skeleton,and designs a cotton yield prediction system based on Android mobile devices.This provides new methods and ideas for cotton yield prediction,and has certain application value.The research content is as follows:(1)A modified YOLOv5x+algorithm was proposed to identify cotton after spraying defoliant on cotton fields as the research site.By embedding convolutional attention modules,using deep separable convolutions,and adding detection layers for small-scale cotton,the detection accuracy of small target cotton was improved.The trained YOLOv5x+model had an average detection time of 78.43 ms per image,The accuracy P(%)and recall R(%)are 90.95 and 89.16,which are 19.58 and 16.84 percentage points higher than the original YOLOv5 x model,respectively.This meets the cotton detection task and has higher accuracy than the original algorithm,making it suitable for the field of cotton detection.(2)In view of the difference in the impact of different soil type on cotton yield,through manual sampling of cotton in different soil type,the sampled cotton was measured for the number of cotton bolls,single boll weight,seed cotton weight,lint percentage,etc.,and the sampling data were analyzed by function fitting.The distribution maps of single boll weight and seed cotton weight of different soil properties and the fitting relationships between the number of cotton bolls per plant and the weight of cotton bolls and seed cotton weight under different soil properties were obtained,The average single boll weight in loam soil is7.93 g,the average single boll seed cotton weight is 5.84 g,the average single boll lint weight is 2.46 g,the average single boll weight in clay is 7.59 g,the average single boll seed cotton weight is 5.34 g,the average single boll lint weight is 2.24 g,the average single boll weight in sandy soil is 6.31 g,the average single boll seed cotton weight is 4.68 g,and the average single boll lint weight is 2.01 g.The established fitting function relationship model between the number of cotton bolls per plant and the number of seed cotton provides data support for the mobile cotton yield prediction system developed based on the Android system.(3)Developing a mobile cotton yield prediction system using the Android platform,with the improved YOLOv5x+algorithm model as the core algorithm and combined with the fitted cotton yield mathematical model,the corresponding APP is developed using the Android Studio development platform.By selecting mobile phone photography or calling a photo album to obtain image information,the target image is analyzed and processed to achieve cotton yield prediction,Using the detection box of cotton in each image,the cotton bolls are detected,and the weight of cotton in different soils is brought in.The cotton yield per hectare is automatically calculated.Compared with the actual yield,the average error between the actual yield per hectare and the predicted yield of seed cotton and lint is 8.134 kg and 3.865 kg,which is not significant and can meet the actual production needs.