Research On Sex Recognition Method Of Silkworm Pupa Based On Machine Vision Technology

China is the world’s major silkworm breeding country and the birthplace of the Silk Road.High-quality silkworm varieties are the key to improving silk yield and quality.Therefore,in order to obtain high-quality silkworm varieties,it is necessary to separate the female silkworm pupae and male silkworm pupae of the same variety first,and then cross-breed the female silkworm pupae and male silkworm pupae of different varieties.The sex recognition and sorting of silkworm pupa is a key link in breeding.The quality of hybrid silkworm species and silk quality is directly affected by the accuracy of the sex separation of silkworm pupa.At present,the sex recognition of silkworm pupa in silkworm breeding is mainly done manually.The labor intensity of workers is high and the sorting efficiency is low.There is an urgent need for intelligent sex recognition of silkworm pupa and automated sorting equipment to replace manual sorting.The development of an accurate and fast automatic sex recognition method of silkworm pupa is the key to realize the automation of silkworm breeding.Machine vision technology has the advantages of high accuracy and losslessness,and has been widely used in the agricultural field.Silkworm pupa is living,and the appearance of female and male silkworm pupa is small.Different gonad shapes are its main characteristics.For small target objects such as silkworm pupa images,there are still many key technologies that need to be studied using machine vision technology to identify male and female silkworm pupas.This article is based on machines.Visual research on the automatic sex recognition method of silkworm pupa.The main research contents are as follows,(1)Research on sex recognition method of YOLOv4 silkworm pupa based on machine vision technology.When constructing the YOLOv4 sex recognition model of silkworm pupa based on machine vision technology,because the pupa is living,the tail swings and the shape is irregular ellipsoid when the machine vision imaging system acquires the silkworm pupa image online,which results in the collection of online data.In addition to the frontal images of the gonads,the image data also has non-frontal images,and neither the side nor the back images contain gonadal features.Discuss the influence of silkworm pupa image data set with different features such as the front,back and two sides of gonads on the accuracy of model recognition.The YOLOv4 sex recognition model with only the frontal image set of gonadal features has an accuracy rate of 84.41%,and the accuracy rate of recognition on verification set was 83.12%.The recognition time of this model for a single image is between 93.394～99.733 ms.The YOLOv4 recognition model self-verified recognition set accuracy rate of the front,back and two side image sets is 68.20%,and the recognition accuracy rate of his verification set was 70.69%.The recognition time of this model for a single image is between93.393～99.365 ms.The test results show that the YOLOv4 recognition model with only the front image of the gonadal features has the highest recognition accuracy,which indicates that the back and left side images of silkworm pupa are not uniform in texture characteristics and similar to some front images,resulting in a decrease in the recognition accuracy of the recognition model.Calculate and compare the gradient feature value and texture complexity of each surface image of silkworm chrysalis,use the gradient feature value and texture complexity to select the gonad-containing frontal image,and the selection accuracy is 75.82% and 52.74% respectively;Using Local Binary Pattern(LBP)and Local Phase Quantization(LPQ)respectively to extract the texture features of the front,side and back images of silkworm pupa,and then combine with support vector machine to construct the frontal image of silkworm pupa with gonads Select model.The LBP feature extraction combined with the support vector machine classification model has the best classification effect.The accuracy rate of the frontal image containing the gonad is 96.81%.It shows that for silkworm pupa images,the rotation invariance and gray-scale invariance of LBP have practical effects.The feature extraction method combined with the classifier can effectively select the frontal image of silkworm pupa containing gonads.(2)For images with gonadal features,explore the impact of different annotation scales on the recognition model.The whole image of silkworm pupa with gonad,tail part of silkworm pupa with gonad and gonad part of silkworm pupa were labeled respectively.The sex recognition model of silkworm pupa was constructed by using yolov4 algorithm,and the sex recognition results of silkworm pupa were compared.The YOLOv4 recognition model with tail features has a recognition accuracy of 97.87%,and the recognition accuracy of the verified is 84.84%,the recognition time of this model for a single image is between 94.516 and 100.268 ms.The YOLOv4 recognition model that annotates the entire silkworm pupae has a recognition accuracy of 79.60%for the self-validation set,and the recognition accuracy of verified is 68.01%.The model’s recognition time for a single image is between 93.633～100.340 ms.The YOLOv4 recognition model for marking silkworm pupa gonads has a recognition accuracy of 80.30% on the self-validation set,the recognition accuracy of verified is59.06%.The recognition time of this model for a single image is between 93.957 and100.755 ms.Experiments have shown that the recognition accuracy of the recognition model with the tail part is higher than the recognition accuracy of the recognition model with the whole silkworm pupa and only the gonadal part,which indicates that the gonad texture feature and the tail geometry feature included in the tail are more accurate for the training of the recognition model..(3)Comparison of machine vision recognition and near-infrared spectroscopy sex recognition models for silkworm pupae.The research team studied the sex recognition of silkworm pupa based on near-external spectroscopy in the early stage,and achieved good results,but there is still the problem of poor transferability of the recognition model.Aiming at the problem of poor transferability of the recognition model,the sex recognition model of silkworm pupa based on near-infrared spectroscopy was first updated,and a model update method based on semi-supervised learning was proposed.The test results show that the recognition model based on the update of the near-infrared spectrum has been greatly improved.The recognition accuracy of the varieties 872A933,Ming 970,and chrysanthemum 475 have been increased from 86.84%,54.55%,and68.52% to 97.37%,100%,and 96.30%,respectively.Aiming at the transferability of the silkworm pupa sex recognition model based on machine vision,a YOLOv4 silkworm pupa sex recognition model based on the characteristics of the gonad-containing tail of the silkworm pupa was constructed,and the self-validation set,the same batch of different species samples and different batches of different species samples were respectively identified.The model’s recognition accuracy of the self-validation set is98.37%,and the recognition accuracy of the same batch of different species test sets is99.39%.The recognition accuracy of different batches of different species test sets is90.83%.The image recognition time is 93.625～99.379 ms.The test results show that the machine vision-based sex recognition model of silkworm pupa has good transferability between different species,which is better than the transferability of near-infrared spectroscopy for male and female recognition of silkworm pupa;the transferability between different batches of different species is relatively weak,but Compared with the sex recognition model of silkworm pupa based on near-infrared spectroscopy,the transferability of the model is still better,and the model does not need to be updated.Therefore,for the research on ses recognition of silkworm pupa,the image can fully reflect its gender characteristics more comprehensively and concretely.The machine vision technology of silkworm pupa sex recognition model has better transferability and is more suitable for application in actual production.