Abstract: Konjac is susceptible to various diseases during the planting process. In order to automatically monitor the konjac disease in real time, the automatic identification algorithm of konjac disease based on machine vision was studied. Taking Inception V3 as the theoretical model of Convolutional Neural Network（CNN） algorithm, under the deep learning development environment, using the neuron structure algorithm, the neural network was built with neurons as the basic unit, and the identification of konjac disease types was realized. The precision and accuracy of recognition were improved through preprocessing such as normalization and refinement, and the internal and results of the model were visualized to increase the practicability of the algorithm. In the process of recognition, the model was optimized by adjusting the parameters and layer structure, so that the model could better balance accuracy and efficiency. The test results showed that the proposed algorithm could realize automatic disease identification of common konjac, and the accuracy rate was kept above 90%.

Key words: konjac disease, Convolutional Neural Network（CNN）, Inception V3, deep learning