Sex Discrimination Study Of Silkworm Pupae Based On Machine Vision Technology

Sericulture industry has a long history in China,which contributes to the financial revenue of country and income of farmers.The quality of silk varies greatly among different species of silkworm pupae.The crossbred silkworm pupae can produce high quality silk.The accuracy of sex classification of silkworm pupae determines the quality of crossbred silkworm pupae.Sex discrimination of silkworm pupae is usually accomplished manually,which makes it tedious,laboriously and costly.In order to automatically differentitate the sex of silkworm pupae with high performance,many related studies have been done.In conclusion,these methods universally need expensive equipments and have the problems of low accuracy and generalization ability of the identification model.In this work,machine vision technology,which has been widely applied in the agriculture,is used to classify the sex of silkworm pupae.The sex discrimination of live silkworm pupae with small volume is much more complex than other static objects.The main contributions are organized as follows,1.The image restoration algorithm is respectively proposed to recover the low-quality silkworm pupa image with low illumination and noise?spatially variant blur and motion blur.In order to improve the quality of degraded silkworm pupae images,a novel method combining Shan Q's tone mapping with Tikhonov regularization,which is capable of enhancing image contrast and compressing noise simultaneously,is proposed.The results show that the performance of the proposed method can improve the quality of images that are degraded by noise and low illumination at the same time.The method is robust to the variation of noise.An effective method is presented that the complicated restoration of spatially-varying blur silkworm pupae image is decomposed into the simple restoration of multiple images which have the same level blur and are part of original image.The experimental results show that the performance of the proposed algorithm is better than Shen's method.The method successfully removes spatially-varying blur and enhances the image quality.To increase the image quality and relieve the difficulty of discrimination due to motion blur in the silkworm pupae image,a restoration strategy including three stages is proposed.In the image prediction stage,the sharp edges of silkworm pupa image are recovered.In the kernel refinement stage,we incorporated the Radon transform to estimate the accurate kernel.In the final latent image restoration step,a TV-L₁ deconvolution model is carried out to render a better deblurred result based on the accurate kernel.The experimental results show that the sharp edges and textures in the silkworm pupae image are increased to the extreme extent.In addition,the method is robust to the variation of motion blur.2.The sex discrimination is expolored using convolutional neural network?CNN?.A CNN model based on LeNet-5 network that is suitable for silkworm pupae image is built,in which the number of convolution layers and the size of convolution kernel are optimized.The training set and prediction set respectively contain 1400 and 633 silkworm pupae images.The experimental results show that the CNN model gives better performance with accuracy of 95.10%,less time costing per image?0.024s?and better model generalization ability when compared with the feature-based method.The proposed image restoration methods all can improve the performance of CNN model.Furthermore,the result of CNN model is up to 97.15%after processed by motion deblurring algorithm.The results reveal that CNN model is suited for sex discrimination and the accuracy is not influenced by the species.3.Hyperspectral imaging?HSI?technology is explored to differentiate the sex of silkworm puape.First of all,the selection of differnet region of interests?ROIs?including head?center and tail is discussed.Then,24 characteristic wavelengths of preprocessed HSI data are chosen by successive projection algorithm?SPA?.Gray-level co-occurrence matrix?GLCM?analysis is implemented on the first principal component?PC?images to extract textural features.Experimental results show that the SVM model based on fusion data gives the highest accuracy of95%,which is better than the results of only using spectral data or textural data.It indicates that HSI technology can be served as a new method to classify the sex of silkworm pupae.4.The silkworm pupae discrimination and sorting equipment based on machine vision technology is developed,which can realize a series of functions including automatic transporting?image acquiring and processing?identification and sorting.The accuracy of online sex discrimination is up to 96.23%.The time from image processing to differentiation of one silkworm pupa is 0.5 second,and sorting time is 0.6 second.The performance of equipment is not influenced by the species,which can further promote the intelligent breeding development of sericulture.