Detection of undesired inclusion in optically translucent fruits or other objects using noninvasive near-infrared technology

Abstract

In this paper, a noninvasive optical method has been proposed for real-time detection and estimation of undesired inclusion or degradation in optically translucent objects. A novel artificial neural network (ANN)-based technique is fused with conventional photometric and tomographic methods to achieve a reliable estimate in real time. The overall optical depth is computed using the Beer–Lambert law and compared with two estimated thresholds. If the value of optical depth is within the thresholds, the location and size of inclusions are estimated by using an ANN-based method. Finally, the model-based iterative method with estimated a priori knowledge is used to reconstruct the optical parameters. Usage of continuous wave (CW) mode instrumentation with a small number of source-detector positions makes the overall system cheap, portable, and computationally efficient. Extensive simulation studies have been carried out for varying size, location, and shape of the patches to analyze the performance. Experimental validation is also performed thoroughly by using phantoms with multiple inclusions and degraded apples, which shows good potential for the proposed method. Thus, the method can be applied to estimate the condition of fruits for quality inspection and sorting applications based on near-infrared imaging.