Data-Matrix二維條碼因其抗損毀能力強、所佔空間小等特性，已被廣泛應用於航太工業、汽車製造業、半導體與印刷電路板零件的識別。條碼偵測是條碼辨識的關鍵。針對低品質的條碼影像，例如扭曲、模糊、光照不均等，典型的條碼偵測方法會有偵測率過低和辨識困難的問題。本研究提出一個創新的條碼偵測方法，基於Data-Matrix二維條碼其邊緣取向具有相互垂直的特性，將條碼影像轉換成紋理能量圖，再結合卷積神經網路進行條碼偵測模型的深度學習。我們使用一個低品質Data-Matrix條碼影像資料庫來驗證所提出的方法，其條碼偵測正確率可以提升22%，可有效改善條碼辨識性能不足的問題。;Data-Matrix 2D barcodes have been widely used in the identification of aerospace industry, automotive industry, semiconductor and printed circuit board parts due to their strong resistance to damage and small space. Barcode detection is the key to barcode recognition. For low-quality barcode images, such as distortion, blur, uneven illumination, etc. The typical barcode detection methods have problems of low detection rate and difficulty in identification. This study proposes an innovative barcode detection method based on Data-Matrix two-dimensional barcode whose edge orientation has mutually perpendicular characteristics, converts the barcode image into a Texture Energy Map, and combines the depth learning of the barcode detection model with the Convolutional Neural Network. We use a low-quality Data-Matrix barcode image database to verify the proposed method. The barcode detection accuracy can be improved by 22%, which can effectively improve the problem of insufficient barcode recognition performance.
