Research On Swarm Intelligence Optimization Algorithm And Its Application In Several Problems Of Smart Farming

With the improvement of people's living standard and the increasing concern about food safety,and the pressure of supply and demand of agricultural products caused by population growth,traditional farming no longer meet the real demand.Smart farming is the integrated and comprehensive application of digital technology in planting systems for information sensing,processing,management decisions and etc.The implementation of smart farming helps optimize the processes of soil suitability evaluation,crop nutrient modeling,yield prediction and disease management decision of farming systems,which can significantly improve the land yield and resource utilization rate of farming systems and promote the development process of agricultural informatization and modernization.Soil,fertilizer and crops are the three important components of smart farming,how to accurately predict the spatial and temporal evolution of soil,how to seek a more accurate and scientific ratio of fertilizer application,and how to accurately segment and identify the disease spot images of crops are the key technologies to solve the comprehensive management of soil erosion,accurate fertilizer application,pest and disease identification and diagnosis in current smart farming.Swarm intelligence optimization algorithms are more effective in dealing with engineering optimization problems and are receiving more and more attention from researchers,combining machine learning,computer vision,pattern recognition and other methods are widely used in smart farming.However,when solving complex optimization problems,swarm intelligence optimization algorithms suffer from premature convergence,easy to fall into local optimum and slow convergence,while traditional machine learning models suffer from low computational accuracy,poor stability and poor adaptability.In order to improve the performance of the algorithm,many researchers have proposed a large number of variants of the swarm intelligence optimization algorithm in order to obtain an optimization algorithm with better performance.At the same time,improved swarm intelligence optimization algorithms are combined with existing methods such as machine learning to complement each other to achieve better solution to specific application problems.In this thesis,the swarm intelligent optimization algorithm investigates the three hot issues of soil erosion classification,scientific fertilizer rationing and disease image segmentation in soil,fertilizer and crop systems.With the goal of improving the accuracy or enhancing the effectiveness,we provide new solutions and new ideas to assist soil erosion management,precision fertilization,and disease diagnosis.The main contribution of this thesis are as follows:(1)We address the current research status of swarm intelligent optimization algorithms and smart farming separately.Firstly,we present the current status of domestic and international research on three smart farming problems(soil erosion classification,precision fertilization,and agricultural disease image segmentation).Then illustrated and analyzed the three types of swarm intelligence optimization algorithms,machine learning classification methods,image processing and pattern recognition and other related techniques used in this thesis and the existing problems.Finally,we present the performance evaluation indexes required for each experiment to provide a theoretical basis for further research.(2)To address the problem of classifying soil erosion in smart farming and improving the accuracy of classification prediction,we propose a moth flame optimization algorithm with sine and cosine function strategy(SMFO).The use of a sine and cosine function strategy increases the diversity of the initial population and accelerates the chance of jumping out of the local optimum.Experimental results comparing on 25 benchmark functions show that the SMFO algorithm has significantly improved in terms of balance and diversity.Tackling the problem that machine learning classifiers are influenced by key parameters in the prediction process,the proposed SMFO algorithm is applied to optimize the penalty parameter c and the kernel parameter ? of the kernel extreme learning machine(KELM)to improve the performance of the classifier.Further,the model was used to predict erosion classification problems caused by rainfall to obtain soil erosion classification prediction results.The experimental results show that the SMFO-KELM method proposed in this thesis has a better prediction results when predicting the erosion,while the effect is higher than other methods in several evaluation indexes.(3)To address the problem of seeking the best fertilizer ratio and predicting maximum yield in smart farming precision fertilization,we propose a multi-strategy improved grey wolf optimization algorithm(SLEGWO).The algorithm adopted four mechanisms: slime mould algorithm(SMA),lévy flight(LF),opposition-based learning(OBL),and greedy select(GS),the use of integrated strategies accelerates the speed of search and convergence,and the comparison results on 30 benchmark functions show that it provides the best performance.In response to the problem of unsatisfactory fitting of fertilizer effect function,the proposed SLEGWO algorithm was applied to the ternary fertilizer effect function of nitrogen,phosphorus and potassium to improve the fitting effect,which combined with the “3414” fertilizer application test scheme,and then was used to solve the equation coefficients of the fertilizer effect equation and to predict the optimal fertilizer application rates and maximum yield,respectively.The experimental results show that the proposed SLEGWO algorithm enhances the fitness of the coefficients of the fertilizer effect equation,and it also provides more accurate fertilizer application ratios and yield estimation.(4)To address the problem of crop disease image segmentation and segmentation effect enhancement in smart farming,we propose a grey wolf elited comprehensive particle swarm algorithm(GCLPSO).The GCLPSO algorithm effectively enhances the local search capability of the algorithm using the gray wolf ranking strategy,and provides the best performance when compared with other algorithms on 40 benchmark functions.Meanwhile,for the two problems of disease image thresholding segmentation: noise interference and difficulty in finding the optimal threshold.On one hand,combining non-local mean filtering to remove the noise in the disease image;on the other hand,using Otsu segmentation as the objective function,the proposed GCLPSO algorithm is used to solve the problem of selecting the optimal segmentation threshold,which valifying and comparing among multiple maize disease images.The experimental results show that the multiple disease image segmentation method of maize based on GCLPSO and Otsu segmentation outperforms other methods in terms of overall segmentation accuracy,and it provides better robustness and stability.