Abstract: Abstract: The existing detection devices for moldy core in apples have critical shortcomings like high detection costs, high energy consumption, complicated analysis processing and redundant spectral information in wide waveband, for the light sources in most of them are composed of high-power and wide-band illuminant (like tungsten-halogen bulbs), and data acquisition and processing of them rely on the spectrometer and the computer. In this paper, a portable and low cost nondestructive detection device for moldy core in apples was designed based on the conclusion that the apple diameters and the intensities of some characteristic transmission spectra were the most important affecting factors of nondestructive detection for moldy core. In order to determine the specific band related to moldy core in apple, a test platform was built to obtain the transmission spectrum data by using tungsten-halogen bulbs, spectrometer and computer. A waveband of 690-730 nm was confirmed as the characteristic spectrum band through correlation analysis from a band range of 200-1025 nm. According to this band, a series of narrow band LEDs (light emitting diode) were adopted and used to make a special light source. Then, an information collection device was designed to obtain apple diameters and intensities of characteristic transmission spectra. The device consisted of a core processing unit, a driver circuit of LEDs, a conversion and detection circuit of photoelectricity and a measure module of apple diameter. Microprogrammed control unit (MCU) MSP430 was used as the core processing unit to call other function modules and process data. Pulse width modulation (PWM) chip PT4115 was used as a driver to control let-through currents of LEDs. The linear guide and limit sensor were combined to measure apple diameters online. To get an intelligent detection model for sorting out moldy core apple based on apple diameter and intensities of characteristic transmission spectra obtained by above information collection device, 120 Fuji apples produced in Qianxian, Shaanxi were selected, which were always put in a refrigeration storage at a steady temperature of 4 ℃. These apples were taken out on May 1, 2015, and after they returned to room temperature, 80% of them, which were placed in the same stalk directions, were used as the test samples to build discrimination model. The statistical data showed that 24 of samples were moldy core apples and the other 72 were healthy ones. A back-propagation (BP) neural network was adopted to set up a dependable discriminant model after analyzing the distribution diagram of samples, the inputs of which were apples' diameters and characteristic spectral transmission intensities, while the output was apples' healthy status. Apples' healthy status only had a positive or negative value: 1 represented a healthy apple and -1 meant a bad one. The detection model was converted to embedded language and downloaded into MCU. Now an integrated and model- embedded device was made successfully. The remaining 20% apples of the total samples were used to test the designed detection device with the embedded intelligent discriminant BP model. All test sets were requested to have the same placement position as the apple position when BP model was built. The test result was that one in 6 sick apples was detected wrongly and all of 18 healthy apples were detected correctly, which showed an accuracy of 95.83%. The phenomenon that this apple was identified wrongly was due to the emptiness and mild moldy core in apple core, for it was easily sorted out when an apple had an obvious symptom. This study indicates that the designed detection device has a good ability to distinguish apples with moldy core strongly and reliably, and it will have a good application prospect in apple industry. Meanwhile the research method for detecting internal disease of apple in this paper also can provide a reference for other fruits.