The Automatic Identification System Of Silkworm Chrysalis Gender Based On Computer Vision

Silk industry is a traditional industry in our country. It plays an important role in national economic construction, it can increase farmers’ income and employment, expand export, enrich the domestic market and increase the fiscal revenue. The accuracy of silkworm chrysalis gender identification is directly related to the quality of the breeding, male and female silkworm chrysalis sorting is an important part to manufacture high quality silkworm eggs. The silkworm egg field uses manually labor to identify pupa, the labor intensity is big and the production efficiency is low, so this paper is studying using of image recognition method for automatic identification of silkworm chrysalis male and female based on the different characteristics of male and female pupa body, aim at providing theoretical basis and the specific implementation method for automated identification of silkworm chrysalis gender. In this paper, the work are carried out as follows:(1) Through consulting relevant literature, we compare and analyze the pupa sex identification methods at home and abroad were reviewed and decide to use computer vision technology, to obtain the male and female silkworm chrysalis’shape feature and texture feature on bottom after the image pre-processing, then use the BP neural network to classify and recognise.(2) We designed the automatic image acquisition device, through placing three cameras which surround the transparent u-shaped slot at120degrees with each other in circumferential direction, we make a image’s comprehensive range acquisition for silkworm chrysalis which in the u-shaped slot, then gain the silkworm chrysalis’true color image, namely the RGB image.(3) We do pre-processing work such as graying, filtering noise, image enhancement, edge detection, threshold segmentation and mathematical morphology to the gained pupa image, then,depending on the difference of the body and tail characteristic we feature extraction, we extract the geometry features such as pupa area, perimeter, circularity, eccentricity and texture features such as energy, contrast, correlation, entropy and inverse difference moment.(4) This paper use BP neural network as the classifier model, choose the pupa area, perimeter, circularity, eccentricity, energy, entropy as input vector and discuss the influence to model prediction accuracy which network layers, node number and the initial model parameters act on, at last determine the most suitable parameters.(5) Using the model to make classification and recognition test on two varieties of silkworm chrysalis--871A and7532which has100samples respectively. Then75samples are randomly selected as the training samples, and the remaining25as the identify samples, the experimental results show that the prediction accuracy reaches96%and92%respectively.In summary, the image recognition method is feasible to identify the gender of the silkworm chrysalis.