[Objective]This paper aims to solve the problem of accurate recognition and localization of cotton with different postures and grades by cotton picker under the requirement of high-quality cotton picking.A cotton detection method YOLOX-Cotton based on the improved YOLOX is proposed.[Methods]YOLOX-Cotton uses YOLOX as the main framework,including a recognition module and a localization module,and incorporates coordinate attention(CA)module and SIoU loss function,and takes various posture and grade cotton pictures as data sets to train and test.[Results]The detection module of YOLOX-Cotton was capable of detecting cotton with different postures and grades,and the model precision,recall and average precision reached 92.9％,86.8％and 92.4％,which were improved by 5.2,5.5 and 6.1 percentage points,compared with the original YOLOX,respectively.The localization module of this model was capable of accurately obtaining the location of the cotton,the measurements were kept within the threshold range of the validated results of the field trial,and the standard deviation of all samples was less than 0.01.[Conclusion]The experiment proves that the YOLOX-Cotton can effectively solve the problem of cotton detection and localization by cotton picker under the requirement of high-quality cotton picking,and provides strong technical support for the realization of high-quality cotton picking.
cottontarget detectionthree dimensional localizationattention mechanismloss function
维普
万方数据
