Detection And Classification Of Subsurface Fresh Bruises In Apples Using Spatial-frequency Domain Imaging

China is the world’s largest apple producer and also the world’s largest apple exporter.In 2021,the total apple production will account for 55% of the global total,and the total export of Chinese apples in2021 will be second only to the total export of the entire European Union region.The mechanical damage caused by apples during harvesting,transportation,packaging,and other processes is invisible at the initial stage of the damage,and its appearance is similar to that of normal apples.It is difficult to identify with the naked eye,and is called subsurface fresh bruises.Spatial frequency domain imaging(SFDI)technology is a new optical detection technology,which can indirectly measure the optical properties of biological tissues in a non-contact,single measurement way.In recent years,the technology had great performance in the quality and safety evaluation of fruits and vegetables.However,there are still many deficiencies in this technology,such as limited optical transmission model and low computational efficiency.The mechanism of detecting apple subsurface fresh bruises based on wide-field optical property distribution is also unclear.Therefore,from the perspective of the interaction between light and tissue,this topic explores the feasibility of SFDI technology for detecting and classifying different apple groups.The main research contents and conclusions are as follows:1.Optimize and upgrade the spatial frequency domain imaging system,and conduct parameter calibration and performance evaluation.The original single wavelength imaging system has been improved to a multispectral imaging system,and system calibration and verification experiments have been conducted on the system,including camera calibration,image uniformity correction,system linearity correction,and system verification.In precision,repeatability,and stability experiments,the magnitude of measurement errors and standard deviations met requirements.2.The light penetration depth in apple tissue was analyzed theoretically and experimentally using SFDI technology.The combination of MC simulation and developed programs shows that the light penetration depth in apples does not exceed 2.2 mm.The actual experiments on USAF 1951 targets covered with apple slices of different thicknesses show that the maximum light penetration depths are 1.8mm(with peel)and 2.3 mm(without peel),which are very similar to the simulation results.The apple damage detection experiment shows that the maximum depth of apple damage should be around 1.2 mm below the surface peel,otherwise SFDI technology cannot achieve accurate detection.3.The classification research of apple subcutaneous recessive injury was carried out.Using the optical characteristic maps of undamaged and damaged apples,two classification models of support vector machine(SVM)and convolutional neural network(CNN)were proposed.SVM was used to classify the two categories(intact,bruised)in real time,with a maximum accuracy of 98.33 %.However,the accuracy of the three categories(intact,mildly-bruised,severely-bruised)is only 82.50 %.After that,the accuracy of two and three classification using CNN was improved to 99.16 % and 91.50 %,respectively.Finally,the optical characteristics of damaged apples were monitored during 24 hours of storage at room temperature.