China’s agricultural products,food production enterprises,small-scale,scattered,uneven quality,market consumers and channels are a great many.The food industry is more and more prosperous today,people have invested a lot of work in the field of food testing.In fruit,apple is one of the most popular fruits in the world,and it has great benefits to human health.The market demand is so great that nearly 400 billion apples are eaten a year,which shows how people love it.In order to maintain the freshness and uniformity of apples,as apples grow and mature from fruit trees until we see in supermarkets on shelves with species labels.In harvest seasons,fruit farmers usually sort apples by eyes and pack by hand,consuming a great deal of manpower and material resources,resulting in low efficiency,reduction of apple quality,and backlog of products.It is therefore necessary to artificial intelligence in this process.The traditional apple classification methods have become mature,but there are still some limitations.Usually this involves the use of complex laboratory environment,large investment,and sometimes tedious experiment process.Thus the traditional methods are not suitable for wide use in real life.In this thesis,a novel study of feature extraction and classifier model for apple classification is reported.The work includes the following two parts: a feature extraction method and a classification method.A feature fusion method for apple classification is proposed,which is based on multi-angle and multi-region feature fusion.Images of the apple is taken from multiple angles,and a plurality of smaller regions from each image are extracted.A color histogram is used to represent each region,and all region histograms are combined to create a single feature vector to represent the apple,which is then used for classification by a machine learning model.Experiments show this method is quite effective,achieving accuracy as high as 97.87%.This apple classification method is simple and cost effective.Based on the feature fusion method of apple classification,a two-layer classification model is proposed and applied to apple classification.The classification model redefines the original training data set in each pair of classes,and creates the corresponding classification model in the base classifier,which is called feature extractor model.Thenthe original training data set is used through these feature extractor models in turn,and the output feature vector is combined as the feature elements of the reconstructed dataset to generate the reconstructed dataset.Finally,A classification model is created on the base classifier with the reconstructed training data set.The classification model is applied to the apple feature data of feature fusion,and better classification results are obtained,which is suitable for conversion into mobile phone applications and promote to the market. |