Statistical methods for the analysis of thermal images obtained from corn seeds

During the last decades, digital image processing algorithms have been developed to measure external characteristics of agricultural products due to the great potential that these methods offer. So, in this research, the thermal images obtained from a thermographic camera were analysed considering two genotypes of maize seeds: crystalline and floury in their natural state, previously irradiated with a laser light source of 650 nm for exposure times of 15 s and 35 s. The methods applied in the analysis were: a) histogram to obtain the distribution of gray levels of images, b) mean value that indicates the brightness of images, c) variance which means the contrast of images, d) entropy applying both Shannon and Tsallis definitions, which provide the average self-information of images, e) estimation of the probability density of temperature variations on seeds to quantitatively characterize them from thermal images. Higher mean and variance were obtained from crystalline seeds indicating higher brightness and contrast. Furthermore, thermal images of floury seeds had higher entropy of Shannon indicating that images had greater disorder with respect to images of crystalline seeds. In the case of the entropy of Tsallis, the entropic index q could be used for characterization of seeds. Thermal images obtained from seeds with a floury structure provided a higher redundancy value for a shorter exposure time to laser light. Thus, the viability of the statistical methods of digital image processing applied to thermal imaging for the characterization of seeds is shown.