No abstract

Early apple bruises, especially those occurring within half an hour, usually have no external symptoms and are di cult to nd. In this study, a fast and nondestructive detection method for early bruises based on a near-infrared camera and image recognition was developed. A total of thirty apple samples were photographed on both sides of each apple. Grayscale images of the apples were captured using a nearinfrared camera with a wavelength region between 900 and 2350 nm. Images of apples (n = 62) without bruises were collected. The same apples were arti cially damaged and photographed by the near-infrared camera immediately. The apples were photographed again at 30-35 min after bruising, and a total of 186 grayscale images were collected. As the glossiness of apples limits the accuracy in the detection of defects, a compound method was proposed consisting of nonlinear grayscale transformation and frequency-domain image ltering techniques, followed by the rst derivative to obtain the gradient grayscale image. Since bruises had distinct edges, bruise edge pixels were detected instead of sound bruise pixels. The compound method obtained a 97.62% classi cation accuracy for nonbruised apples and apples with fresh bruises. The experimental results show that it is feasible to identify early bruises in apples based on near-infrared camera imaging and gradient grayscale images. The method can also provide a reference for the in-situ nondestructive early bruise detection of apples and other fruits.

Early apple bruises, especially those occurring within half an hour, usually have no external symptoms and are difficult to find. In this study, a fast and nondestructive detection method for early bruises based on a near-infrared camera and image recognition was developed. A total of thirty apple samples were photographed on both sides of each apple. Grayscale images of the apples were captured using a near-infrared camera with a wavelength region between 900 and 2350 nm. Images of apples (n = 62) without bruises were collected. The same apples were artificially damaged and photographed by the near-infrared camera immediately. The apples were photographed again at 30–35 min after bruising, and a total of 186 grayscale images were collected. As the glossiness of apples limits the accuracy in the detection of defects, a compound method was proposed consisting of nonlinear grayscale transformation and frequency-domain image filtering techniques, followed by the first derivative to obtain the gradient grayscale image. Since bruises had distinct edges, bruise edge pixels were detected instead of sound bruise pixels. The compound method obtained a 97.62% classification accuracy for nonbruised apples and apples with fresh bruises. The experimental results show that it is feasible to identify early bruises in apples based on near-infrared camera imaging and gradient grayscale images. The method can also provide a reference for the in-situ nondestructive early bruise detection of apples and other fruits.

Determining the homogeneity of material mixing in real time during product processing is critical for quality control. According to the Kubelka–Munk (K-M) function of diffuse reflectance absorption spectrum, absorbance (A) is approximately linear with the content of the components when the sample scattering coefficient (S) is in a certain range. The S is determined by the particle size of powder samples. Therefore, this study determined particle size ranges that satisfy linear additivity in near-infrared diffuse reflectance spectroscopy (NIRDRS). Thus, the proposed NIRDRS analysis technique can be used to determine the homogeneity of material mixes or analyze the percentages of the components in the mixture. In this study, vitamin B3 and vitamin C were used for preparing mixed samples with varying percentages. The experimental results revealed that linear additivity is satisfied when the powder particle size is in the range of less than 280, 280–450, and 450–900 μm. When the confidence level is 0.01, the actual mixed spectra are not significantly different from the “simulated mixed spectra” constructed by linear addition, with their relative deviations less than 1.08%. The absolute errors of the actual and analytic percentages were within 2.98% for each component in the mixtures. The above conclusions also hold for sorghum, which has a complex material composition. Statistical models cannot analyze the percentages of components in the mixture. In contrast, linear addition and direct calibration approach avoids the use of a large number of samples for statistical modeling and analyze the percentages of mixed samples. Meanwhile, it can be used to discriminate and analyze the material mixing uniformity by building a mechanistic model.