Abstract: With the rapid development of agricultural information intelligence, the traditional mode of relying on the sky for food has evolved into a new agricultural mode in which intelligent equipment monitors the production process and performs automatic regulation. How to process the data collected by these smart devices, especially the image data, extract useful information from it, and combine it with the actual production control measures, so as to promote the efficient and rapid operation of agricultural production, of great significance.
This study applies image processing and machine learning to automatic disease identification and leaf age detection of rice, which is more efficient and accurate than traditional manual identification and diagnosis methods.
The main contents include: First, the basic theories of machine learning algorithms and image processing are introduced, and corresponding theoretical guidance and technical solutions are provided for the subsequent research on disease identification algorithms and leaf age diagnosis algorithms.
Then, in the research on the algorithm of rice disease identification, the machine learning method is used for rice disease identification. This paper mainly studies three common diseases of rice, namely rice blast, bacterial blight and bacterial leaf spot.
The specific contents include: First, preprocess the rice disease images, then segment the rice disease spots, and establish the corresponding rice disease image set.
Then, according to the pathological external manifestations of different lesions, multiple features were extracted, and the extracted features were optimized by principal component analysis.
Afterwards, BP neural network and support vector machine were used to establish models, and the optimized features were classified and tested, and the BP neural network with higher recognition accuracy was selected as the final classification model.
Finally, the improvement of the BP classification model is proposed, and the genetic algorithm is used to optimize the initial selection process of the weights and thresholds in the BP algorithm. The experimental results show that the improved GA-BP algorithm is feasible and improves the accuracy of disease identification.
Then, in the research of the algorithm for diagnosing the leaf age of rice, the traditional image processing method was used to extract the veins of the rice leaves.
The judgment method is used to extract the rice leaf veins, and the improved leaf vein extraction algorithm can successfully extract the main leaf veins of rice. And the algorithm was used to verify the diseased rice leaves and the disease-free rice leaves respectively, and the algorithm extraction effect was good. At the same time, based on the algorithm, combined with the leaf vein bias method, the leaf age diagnosis was realized through practical tests.
Finally, combined with the results of rice disease
identification and rice leaf age diagnosis, on the basis of identifying
diseases, taking leaf age diagnosis technology as the core, aiming at different
diseases, and then taking corresponding management measures to paddy fields
according to different leaf age stages in time to carry out Smart regulation of
rice in cold regions.
Chinese Abstract: With the improvement of computing power of computers and chips, research in the field of artificial intelligence has re-emerged. Among them, Convolutional Neural Network (CNN) is widely used in image recognition in various fields, and is gradually becoming more and more mobile and embedded.
Rice disease detection has always been the primary problem in plant disease detection. How to accurately and efficiently detect rice diseases and improve rice yield is a hot issue in agriculture.
Based on artificial intelligence theory, the rice disease detection system is studied. The system is mainly designed from three parts: image acquisition, temperature and humidity acquisition, and disease detection model.
The ARM is used as the main control unit of the embedded platform to realize the image acquisition of rice disease spots and the detection of environmental temperature and humidity. The collected images are used as the input of the detection model to complete the offline detection of rice diseases, and the diagnosis is assisted by environmental factors such as temperature and relative humidity.
Model design on the PC, build the TensorFlow deep learning framework, complete the design and training of the rice disease detection model through the convolutional neural network, and transplant the optimized model to the embedded terminal to realize the detection and recognition function of the terminal.
The test shows that the rice disease detection system can complete
the basic functions. The image acquisition rate is 25 frames per second, and
the resolution of the acquired disease spot image is 640×480. The humidity range
measured by the system is 0-99.9% RH, and the temperature range is -40-80℃. Through test verification, the average accuracy of disease identification is 96%.
Post a Comment