An Improved Neural Network Method for Plant Leaf Recognition

Abstract

Under the background of the general plan for general survey of Germplasm Resources issued by the State Forestry and grassland administration, in order to establish the forest and grass germplasm resources management system, it is urgent to identify the forest vegetation types. In the past, manual identification was often used, which was inefficient and had a high error rate. Considering the small number of samples in the plant leaf database, in order to improve the accuracy of plant leaf identification, In the framework of tensorflow, a convolution neural network method for plant leaf image recognition based on transfer learning is proposed. Firstly, the plant leaf image is preprocessed, and the plant leaf image data set is expanded through the horizontal transformation, random clipping, translation transformation, color and illumination transformation of the original image, and is divided into training set and test set in the ratio of 7:3, the concept V3 model is applied to image data processing by migration learning. The trained models ResNet50 and Concept V3 are migrated and trained on the plant leaf image data set, and the full connection layer is replaced in the pre training model, so that it can adapt to the recognition of plant leaf images. The accuracy of the test set obtained from the pre training model of this method is 95.22% and 95.45%, reaching the excellent level required by the task.