The Study Of Semantic Annotation Technique Of Insects Image

As the research of objects in entomology is deepened and expanded unceasingly, there are more and more insect images made in entomology fields, resulting in the dilation of insects database, which makes the researchers difficult to query and retrieve their needed images information conveniently, fast and accurately. So it is a research hot spot to study annotation of insect images, which is of important value. And in this paper, the annotations of the lepidoptera, coleoptera and orthoptera insect images were studied deeply. The main work of this paper shows as follows: (1) The preprocessing mode of the three classes images were designed.It was finished by smoothing filtering, background removal and the process of isolated noise points. According to the experimental results, this processing not only could finish the image preprocessing but also could segment the objects from the background successfully.(2) Extracting shape features of the somas of coleoptera and orthoptera insects images. And the main operations including:the morphologic dilation and corrosion algorithm processing, the edge detection of Canny and the extraction of global shape feature finally. And it was proved by the experiments that with the more accurate soma contour the global shape feature could be extracted better except a few special ones.(3) Extracting texture features of the three classes objects. And the process was finished by calculating the eigenvalue of the co-occurrence matrix for the three classes images on the basis of background removal. So the experiments results showed that if the images are of fixed size, the extraction could be achieved quickly, or the time complexity would increase rapidly and the delay would also become longer markedly.(4)Researching the implementation principles of neural network algorithm and SVM support vector machine method, analying and comparing the superiority and inferiority of them, and the final classification was achieved using the latter, by which the classification accuracy of the three classes images reached 85% and 88.33% of the later two.(5) The annotation of the three insect images on order. And it was finished by the mapping technique of database tables on the basis of the classification results, achieving 95% above of the recall ratio and 86% above of the precision.(6) Families annotation of locust. It was implemented by using XML language to organize morphological characteristics of the orthoptera locusts, so that the computer could reason out the family automatically which the current object belongs to after attaining the features manually. And both the recall ratio and the precision reached above 80%.