The Feature Extraction And Recognition Of Spore Images Based On Machine Learning

The 21th century is the era of information science which is developing with high speed, and is also the era of life science development, the integration of life science and information science has not only improved the development themselves, but also accelerated the emerge of new study fields. The development of information science has provided a series of convenient methods and tools for life science, so the combination of information technology and biotechnology is becoming a hot research field in recent years. With the rapidly development of modern computer technology, the recognition and classification of image based on artificial intelligence has been more and more widely applied in the automatic detection of pathogenic microorganism in agriculture. Combined with the traditional detection method through microscope, it has many advantages, for example, small subjective error, high working efficient and the required technology standard for the surveyor is not very high, it can significantly increase the effectiveness and accuracy of the forecast of agricultural diseases to satisfy the requirement of agricultural production.Machine Learning is the learning process to simulate human through computers, with the new algorithms are put forward constantly in this field, it is widely applied in image classification and recognition. In this paper, we use the relevant algorithms of Machine Learning to deeply study the classification and recognition of spore microscopy images. The recognition through microscope artificially is widely used in the traditional method of classification and recognition of spore microscopy images. It has high technical requirements for the surveyors and also hard to operate. In this paper, we extract features of the spore images based on SIFT features and RGB SIFT features, and cluster the features we get, so each image can be represented by a vector. This algorithm uses the thought of bag of words, it regards each document as a vector of word frequency. Each document can be regarded as a probability distribution composed of some themes, and each theme can be regarded as a probability distribution composed by many words. In the classification of spore microscopy images, we use feature vector instead of word frequency vector and transform images into different kinds of probability distributions. On this basis, we use different kinds of classifiers to classify the vectors we get, thus we get the results of classification of spore microscopy images. This method gets comparatively good results in feature extraction and classification of spore microscopy images, and it provides theoretical foundation in the forecast and classification of different kinds of spore diseases in agricultural.In order to predict and analysis the different features in the whole procedure of the spore diseases, it is significantly important to recognize the different features of the spores in their life cycle. So in the study of fungus disease, it is necessary to recognize and count the spores in the microscopic image ahead of schedule, and then obtain the survivability and survival rate etc. of spores in different environments, namely adverse situation or favorable circumstances. In this paper, we use convolutional neural network to recognize the spores in the microscopic images, convolutional neural network is a new artificial neural network method that combined deep learning and artificial neural network, compared with other methods it has many advantages, such as local perception area, hierarchical structure, combination of feature extraction and classification and so on. It is widely used in many fields, for example, face detection, documents analysis, voice detection and vehicle license plate recognition etc. In this paper, on the basis of summarizing the basic concept, algorithm and research achievements of convolutional neural network, we recognize the spores in the microscopic image through convolutional neural network by adjust the model and parameters, and we obtain comparatively good results, it provides foundations and convenience in further study of biological research and also provides some theoretical foundation for automation management in agricultural protection in our country.