From experiments to realizations: Hyperspectral systems

Expectations on sorting quality have increased tremendously over the last two decades, regardless of what has to be sorted: minerals, food, or waste. Although the drivers might depend on the type of material, the goal is always the same: companies want to or have to increase the sorting quality while keeping or increasing the throughput. Thus, modern fast optical inspection systems based on visual, e.g., RGB line cameras, are integrated into many production lines. For about a decade, near-infrared sensors are also used for sorting but still play a minor role compared to the huge amount of sensors working in the visible range. The advantage of the near infrared range is that light is reflected which physically interacts differently with the material than electromagnetic waves do in the visible range. Thus, the material can be characterized differently in addition to an RGB-sensor. The disadvantage of a near-infrared sensor is that it is expensive, has got less pixels, and no multi-line sensor exists so far, to mention only a few aspects. State-of-the-art near-infrared sensors are sensitive over a wide spectral range. For discriminating desired product and foreign material it is usually necessary to filter this spectral range. Hyperspectral imaging systems are ideal tools for analysis of material that has to be sorted as they acquire a complete spectrum for each single pixel instead of only three spectral values as an RGB sensor does. Although it is preferable for laboratory analysis it is the exception to the rule to put a hyperspectral imaging system into work for, e.g., mineral sorting, recycling, or food sorting, due to its slowness, expensiveness, and insensitivity.