Peanut kernel integrity detection based on deep learning convolution neural network

Abstract

Objective: To accurately distinguish intact peanut, nut damaged peanut and epidermis damaged peanut. Methods: A peanut seed integrity detection scheme based on deep learning convolution neural network （CNN） was proposed. The peanut seed color selection system was established and a peanut seed image database was also established; The improved density peak clustering （DPC） algorithm was used to adaptively compress the CNN convolution kernel to effectively balance the network depth and operation efficiency; The improved sparrow search algorithm was used to optimize the CNN super parameter configuration and network structure, and the CNN model suitable for peanut grain integrity detection was obtained. Results: Compared with other detection methods, this scheme improved the recognition accuracy by about 5.41%~13.92%, and the detection time of single image of peanut grain was shortened by about 16.9%. Conclusion: This method effectively improves the accuracy and real-time of peanut grain integrity detection.

Keywords

• peanut seeds
• image processing
• convolutional neural network
• density peak clustering
• sparrow search algorithm
• accuracy