Research On Surface Defect Detection Method Of Delinting Cottonseed Based On Machine Vision

The cotton industry has now developed into one of the pillar industries of the southern Xinjiang economy.Since the implementation of the "seed project" strategy,the cotton industry in Xinjiang,especially in southern Xinjiang,has developed rapidly.Therefore,increasing the cotton production in the region is of great importance to promoting the economic development of the region.The emergence rate of cotton directly affects the yield of cotton.The quality of delinted cottonseeds is one of the main reasons for affecting the emergence rate of cotton.Whether the seeds are damaged or not is one of the direct indicators to evaluate the quality of the seeds.The yield can be increased by up to 40%.Therefore,the development of a rapid,nondestructive and accurate method for the detection of surface defects in delinted cottonseeds is of great significance for improving cotton yields and economic development in the region.According to the actual production needs,this paper takes Xinjiang long-staple cotton "Xinhai-63" delinted cottonseed as the research object,and combines the current non-destructive testing widely used machine vision technology to develop an accurate and rapid delinting cottonseed defect.The detection methods,specific research contents and results are as follows:(1)Design and build a delinting cottonseed image acquisition platform.The selection of the industrial camera,lens and light source of the acquisition platform was introduced respectively,and the acquisition of the delinting cottonseed image of " Xinhai-63" was realized by using the acquisition platform.(2)A method for damage detection of delinting cottonseed based on connected regions is studied.First,the collected multi-grain cottonseed images are automatically extracted by using the HSV spatial transformation method,the maximum inter-class variance method and the connected area denoising method;Secondly,the obtained single cottonseed image is enhanced by wavelet denoising with improved threshold;Then,extract the features of damaged areas by applying custom binarization threshold,multiplication,connected area denoising and erosion to the enhanced image;Finally,the method of obtaining the number of connected areas is used to realize the identification of intact cottonseed and damaged cottonseed,and the algorithm is designed as a user-friendly UI interface.The test found that the method based on the connected area discrimination can effectively identify the damaged cottonseed,the result shows,The average accuracy of this algorithm for damaged cottonseed is 89%.It outperforms the soft-threshold function,the hardthreshold function,and the soft-hard threshold compromise function with an average accuracy of 83.5%,85%,and 87.5%.(3)The damage detection of delinting cottonseed based on the improved YOLOv5 s model is studied.First,the original dataset is augmented with darkening,mirroring and rotation methods;Secondly,CBAMNet is embedded in the backbone network to improve the model to obtain the characteristics of cottonseed with less damage;Then,change the original bounding box loss function GIo U to CIo U to make the predicted box regression more accurate;Finally,a detection layer is added to the neck of the original network to increase the model’s ability to learn more dense features.The results show that the overall performance of the improved YOLOv5 s algorithm is better than the current mainstream detection algorithms,and the m AP reaches 94.6%,which is 4.3% higher than the original YOLOv5 s algorithm.Prove the feasibility of the improved model,and the improved model can detect an image in only 0.12 s.The real-time performance of the model is proved.The established traditional image processing algorithm based on connected regions has an accuracy rate of 89% for the detection of damaged cottonseeds,The m AP of the convolutional neural network algorithm based on the improved YOLOv5 s model for the detection of damaged cottonseed is 94.6%.