農業部-水產試驗所全球資訊網農業部-水產試驗所全球資訊網

本研究做出一套魚類影像辨識系統，利用魚類特徵擷取的方法，結合深度學習神經 網路，用以辨識魚市場常見魚類照片。我們使用訓練資料庫的41種魚類1025張魚類 照片，訓練神經網路分類器後，對測試資料庫41種魚類共205張魚類照片進行測試。 系統的平均辨識率為84%，平均每張影像辨識時間為0.95秒。實驗結果證實，我們提 出的方法能夠接受具有複雜背景的魚類影像，而且具有良好的辨識性能和計算速度 。

This study produced a fish image identification system that uses a fish feature extraction method combined with a deep learning neural network to identify common fish photographs in the fish market. We used 1025 fish photos from 41 fish species in the training database, trained the neural network classifier, and tested 205 fish photos of 41 species of fish in the test database. The average recognition rate of the system is 84%, and the average image recognition time is 0.95 seconds. The experimental results confirm that the proposed method can accept fish images with complex backgrounds, and has good recognition performance and calculation speed.