Soil Type Identification And Probability Based On Physicochemical Characteristics

Soil Taxonomy(ST)and Chinese Soil Taxonomy(CST)have been constantly improved,they are theoretical basis for automatic soil-type retrieval.Field Guidelines for Describing and Sampling Soils is a standard for the semantic specification of soil characteristics in China.In recent years,some progress has been made on the research of Soil type retrieval system based on ST or CST,these retrieval systems mainly focus on the integrity of soil data,but still have disadvantages such as follows:1)The coupling of retrieval framework(reasoning process)and CST objects(diagnostic objects and soil types)is neglected,and it impacts on system updates.2)Failure to consider the division of soil information carrier in spatial structure and it's not conducive to the management of soil information.3)The logical relationship between soil characteristics is expressed through nested conditional statements,which makes the retrieval language cumbersome and redundant.Therefore,it is essential to refactor a new retrieval model to remedy the above defects.In this study,the concept of ontology was introduced into model design.Based on the theory of soil geography and CST knowledge,the soil space structure and the logical relationship between soil entity and CST objects were considered,and then the soil ontology model and CST ontology model were designed.To normalize the description of soil characteristics,the soil attribute model is defined,and the soil characteristic types are divided into single characteristic type and compound characteristic type.Moreover,predicate logics were defined as interfaces in order to normatively express the logical relations between types and membership relations between models.The construction of ontology models and the definition of predicate logic are finished by using Python,and the retrieval system is completed with four levels(Order,Suboorder,group and subgroup).Finally,this system was tested by using the representative pedons data in Soil Series of China(Hubei volume).The retrieval result not only gives soil types to the tested pedon at higher category,but also records the whole retrieval process for the results analysis.Compared with other existing retrieval models,this model divides the carriers of soil information into horizon,profile,pedon and polypedon,which facilitates the management of soil information and reduces the complexity of classification rules expression.Besides,the model separates the rules from the framework,it has a characteristic of high cohesion and low coupling and can support the update and extension of the retrieval system better.In the retrieval framework,soil characteristics are no longer used as retrieval objects,but are encapsulated in soil entities,diagnostic objects and soil types to elevate the retrieval objects to the level of category,which is more consistent with human cognition.In the above research process,it is found that the retrieval system involves a large number of soil characteristics,it is very strict with the test data,and the retrieval process should be carried out in strict sequence.However,the prefix words of soil type name in CST(including the first word of sub orders and groups,the adjective of subgroups)have quantitative semantics,and are mapped with diagnostic objects.Therefore,the further research attempts to estimate soil type probability of each profile through the prefix words distribution in the CST.The prefix words in CST are regarded as feature vectors,soil types are regarded as parameters to be estimated,the most likely types of each profile are estimated by the maximum likelihood method,and then based on bayes theory,a set of probabilities of profile belonging to soil types are given.This thesis evaluates the classification uncertainty from the perspectives of confusion matrix and fuzzy similarity.In the confusion matrix,the classification errors on the scale of types are evaluated by the four indexes including User's Accuracy,Produce's Accuracy,Hellden Accuracy and Short's Accuracy,and the classification errors on the scale of overall are evaluated by Overall Accuracy and Kappa coefficient.From the perspective of fuzzy similarity,the classification uncertainty of each type was evaluated by comparing the similarity between the fuzzy membership set of the reference samples and the fuzzy membership set of classified samples(FA_S).The classification uncertainty at each category(Order,Suborder,Group and Subgroup)was evaluated by FOAs.The classification results and accuracy evaluation show that the overall accuracy and Kappa coefficient at four categories are all above 0.7.The FOA_S is above 0.75,that is,the reference samples and the classified samples have a high consistency.