A Zebra Chip (ZC) disease detection system was developed based on hyperspectral imaging (HSI) to minimise economic losses in the New Zealand potato chip industry. Current detection methods for other than heavily diseased tubers require peeling or cutting of potato tubers. A rapid and non-destructive grading method would be ideal to remove ZC diseased potatoes at line before processing. The spectral signatures from a large population (n = 3352) of commercially sourced potatoes were collected using HSI in the spectral range of 550 nm–1700 nm. Spectral signatures of each potato (i.e. 1767 ZC infected and 1585 healthy potatoes) were extracted by segmentation and morphological operations. A calibration dataset (80% of the total population was randomly selected), with and without pre-processing, was used for modelling using the partial least squares discriminant analysis (PLS-DA). The model performance shows 92% accuracy for ZC potato identification on validation data (20% of total population). Waveband optimisation by variable importance in projection (VIP) method revealed 34 wavebands sensitive to ZC diseased potatoes. This optimum set of wavebands allowed ZC identification with 89% accuracy. The experiments demonstrate the potential of HSI for identification of ZC infected potatoes in whole tuber before processing. Efficient removal of diseased tubers would reduce processing losses and provide a potential opportunity to access export markets for intact tubers.

中文翻译：

---

用于识别马铃薯斑马片病的高光谱成像

Zebra Chip (ZC) 病害检测系统是基于高光谱成像 (HSI) 开发的，旨在最大限度地减少新西兰薯片行业的经济损失。目前对重病块茎以外的检测方法需要对马铃薯块茎进行去皮或切割。一种快速且无损的分级方法是在加工前在线去除 ZC 病马铃薯的理想方法。使用 HSI 在 550 nm–1700 nm 的光谱范围内收集来自大量商业采购马铃薯（n = 3352）的光谱特征。通过分割和形态学操作提取每个马铃薯（即 1767 个 ZC 感染和 1585 个健康马铃薯）的光谱特征。一个校准数据集（总人口的 80% 是随机选择的），有和没有预处理，用于使用偏最小二乘判别分析 (PLS-DA) 进行建模。模型性能显示，对验证数据（占总人口的 20%）进行 ZC 马铃薯识别的准确度为 92%。通过投影中变量重要性（VIP）方法进行的波段优化揭示了对 ZC 病马铃薯敏感的 34 个波段。这组最佳波段允许以 89% 的准确度进行 ZC 识别。实验证明了 HSI 在加工前鉴定整个块茎中受 ZC 感染的马铃薯的潜力。有效去除患病块茎将减少加工损失，并提供进入完整块茎出口市场的潜在机会。通过投影中变量重要性（VIP）方法进行的波段优化揭示了对 ZC 病马铃薯敏感的 34 个波段。这组最佳波段允许以 89% 的准确度进行 ZC 识别。实验证明了 HSI 在加工前鉴定整个块茎中受 ZC 感染的马铃薯的潜力。有效去除患病块茎将减少加工损失，并提供进入完整块茎出口市场的潜在机会。通过投影中变量重要性（VIP）方法进行的波段优化揭示了对 ZC 病马铃薯敏感的 34 个波段。这组最佳波段允许以 89% 的准确度进行 ZC 识别。实验证明了 HSI 在加工前鉴定整个块茎中受 ZC 感染的马铃薯的潜力。有效去除患病块茎将减少加工损失，并提供进入完整块茎出口市场的潜在机会。