Research On Betel Core Removal Detection Based On Machine Vision Perception

The removing of the core is an important part of the betel nut processing process.The betel nut shape is irregular due to dehydration and extrusion deformation during processing.At the same time,due to the different water content,the betel nut cores after incision vary greatly.It is very difficult to automatically remove the core from the machine,so the betel nut processing enterprises still rely on manual removal of the kernel.As competition in the betel nut processing industry intensifies,labor costs are getting higher and higher,and manual denuclearization is losing competitiveness.Therefore,betel nut processing enterprises urgently need an intelligent and efficient betel nut denuclearization scheme to enhance their competitive advantage.This paper proposes a machine vision-based solution for the betel nut removal kernel problem.The paper work is divided into four parts:1)Aiming at the problem of betel nut kernel contour detection in betel nut removal kernel process,a betel nut kernel contour detection method based on semantic segmentation was proposed.The segmentation model is based on the VGG16 network,and the fully connected layer is replaced by the convolution layer.The jump structure is added.The shallow features are sampled and then merged with the deep features at the same scale,and the conventional convolution is replaced with the dilated convolution.The product,which reduces the learning parameters,improves the real-time performance of the segmentation model,and finally obtains the FCN-Dilated-VGG-8s segmentation model.The model shows good robustness and achieves accurate and fast segmentation of betel nut images.The accuracy of betel nut image segmentation is 98.79%,single image segmentation is only 0.071s,and the model size is only 37.5%of the FCN-VGG-8s model.By performing edge extraction on the segmented image,a complete and smooth betel nut kernel outline can be obtained.2)The color and texture of the betel nut residue are very different,and the betel nut residue is closely attached to the fruit wall,so it is difficult to accurately distinguish the residue by the conventional algorithm.This paper introduces the idea of semantic segmentation,adjusts the loss function of the betel nut kernel detection semantic segmentation model,narrows the region of interest,segmentes the residual block,and counts the number of residual block pixels in the segmentation result to achieve residual discrimination.The experimental accuracy rate reaches 98.7%.,verify the effectiveness of the algorithm.3)In order to improve the success rate of the betel nut removal core,it is necessary to find the best position for the beak to remove the core.The best position of the lower knife is the end point of the beak core contour head.Therefore,it is necessary to determine whether it is the head while finding the contour of the betel nut kernel,so as to accurately locate the optimal lower knife position for the betel nut.For the head-to-tail discriminant problem,we conducted experiments based on the method of deep neural network.The discriminant accuracy is close to 100%.The algorithm is fast and robust,which solves the problem of betel nut head and tail.4)A betel nut removal kernel test system based on machine vision perception was designed.The system initially realized betel nut kernel contour detection,head and tail discriminant and residual discriminant function,which satisfies the requirements of algorithm verification and method verification.In this paper,the paper studies the technical problems in the process of removing the kernel from betel nut,and solves the problem of betel nut kernel contour detection,betel nut kernel residue discrimination,and betel nut head and tail identification.This provides technical support for the automatic removal of the betel nut from the machine and is of great value in promoting the development of the betel nut processing industry.