本論文では，水産資源量調査のために撮影された海底画像群からホタテの自動計測システムの開発を目的とし，砂場環境に適したホタテ領域の抽出手法について提案する．対象となる砂場環境下では，ホタテは砂に身を隠している．このため，ホタテの殻はほとんど確認することができず，視覚的特徴が少ない条件下で認識する必要がある．そこで，新たにホタテが呼吸することで露出している殻緑領域に特化した局所特徴量を提案し．その結果について検討し，その有効性について述べる．%This paper presents a new method for learning kernel classifiers. First, we formulate a novel learning scheme called "General Loss Minimization (GLM)." The formulation is based on Bayes decision theory which can handle various losses as well as prior probabilities. Then, we propose a new learning method for kernel classifiers derived from GLM. We also show that support vector machines (SVM) can be derived from GLM as a special case. The derivation clarifies some interesting similarities and differences between SVM and the proposed method. Finally, we confirm effectiveness of the proposed method in expriments with artificial and real databases. The exprimental results show that the proposed method achieves almost the same or better accuracy than SVM in spite of stronger sparsity of classifier parameters.
展开▼
