Research On Plant Leaf Images Identification Algorithm Based On Deep Learning

In recent years, the technology of plant identification based on image analysis has become a hot research topic in plant information science. The classification and recognition of plants generally depend on the shape, texture, color and other characteristics of the plant. According to the morphological characteristics of flower, fruit and leaf, the classification is done. The texture, color and morphological structure of plant leaves are different, which is the main basis for distinguishing plant species. The plant leaf image acquisition is convenient, the survival time is long, and the four seasons change is distinct, through the leaf to the plant to carry on the classification research to become the research hot spot of many scholars at present.Based on the depth learning algorithm of convolutional neural network, we can improve the efficiency of image recognition by reducing the feature of the leaf, and can eliminate the noise of the complex background image. The algorithm based on the convolution neural network constructed a eight layer depth study leaf recognition system, and the use of Pl@antNet leaf base and autonomous expansion of plant leaf data to the training sample data, complete testing recognition rate. In order to improve the recognition rate, leaf images of single background and complex background are given the different image pre-treatment schemes. Analysis verify the validity of the algorithm by performing a comparison based on the depth of learning and recognition system based on image recognition systems and leaf SIFT features of multi-classifier recognition system. For CNN+SVM and CNN+Softmax recognition system, the simple background leaf recognition rate could reach 91.11% and 90.90%, and the recognition rate of complex background could reach 34.38%.Implementation of this eight layers depth learning recognition system still has room for improvement, the layers parameter defaults to take, which is lack optimization. At the same time, the image segmentation process can still be regarded as one of the key research in the future. The recognition rate of complex background leaf image is less than 40%, and the improvement of the space is still very large. Leaf which is too similar to the plant classification and identification will be big challenges in future.