Study On Recognition Methods For Stems Of Tomato Plants At Night Active Lighting

Fruit and vegetable production is time-consuming and laborious.In today’s severe population aging,the research and development of fruit and vegetable production robots is becoming more and more urgent and has become a hot spot in the field of agricultural robot research.The operation time of fruit and vegetable production robot can be extended and its operation efficiency can be improved by carrying out fruit and vegetable production operation under the conditions of night.In addition,in order to realize the automation of multiple fruit and vegetable production processes,such as pruning,target spraying,fertilization and picking,the visual system of fruit and vegetable production robot needs to realize the automatic identification of fruit and vegetable plant stems.For this reason,this paper studied the stem recognition method of tomato plants under the condition of night active lighting.The main research contents,test results and conclusions are as follows:(1)Evaluation method for optimal image segmentation results of tomato plants under night active lighting.PCNN(Pulse coupled neural network)model can obtain a series of image segmentation images through multiple iterations,and the optimal segmentation result is usually selected by a single entropy value.Through the correlation analysis of the commonly used entropy evaluation methods,the four applicable evaluation indexes are determined to be the two-dimensional Tsallis entropy,maximum entropy,joint entropy and contrast.Based on the logistic regression model,these four entropy values are combined to form a composite entropy index.The experimental results under LED 0.5w active lighting environment showed that: based on the evaluation method of compound entropy value,the accuracy rate of the evaluation of the optimal segmentation result was 83.34%.The experimental results were compared with the traditional entropy evaluation,and the comparison results verified the effectiveness of the PCNN model image segmentation evaluation method based on composite entropy.(2)Adaptive setting method of parameters of tomato plant PCNN image segmentation model based on night active lighting.Based on the traditional PCNN model parameters set trival,that the problem of poor real-time performance more iterative times,put forward the improved PCNN model,combining with the Otsu algorithm automation parameter Settings of PCNN model,and the links in the PCNN model input item is improved,and ultimately improved algorithm based on tomato plants at night in the process of image segmentation,the need for manual experience value,average segmentation accuracy is 90.43%,superior to Otsu;The average segmentation time of each image is 0.9944 s,which is better than traditional PCNN algorithm,but longer than Otsu algorithm.The results show that the improved PCNN model can improve the segmentation accuracy of foreground pixels while meeting the real time production requirements.(3)Stem recognition algorithm of tomato plants under night active lighting based on Mask RCNN deep learning model.Mask RCNN is a very flexible framework of deep learning,can increase the different branches to complete various tasks,with the advantages of high precision,the improved PCNN model the image after segmentation,based on the Mask RCNN model of tomato plant stalks the night image recognition,40 for testing of tomato night vision image experiment results show that: the tomato plant main stem recall rate was 82.5%,the level of stem the recall rate was 65%;The average recall rate was 73.75%;The model test output AP was 0.421,the m AP was 0.315,and the average test time for each image was about0.15 s.Therefore,the stem recognition of tomato plants could be realized quickly under the condition of active lighting at night.The above research results provide reference for tomato plant stalk recognition under night image segmentation and night active lighting,and lay a certain technical foundation,which can help to extend the working time of fruit and vegetable production robot and promote the research process of multi-link automation in fruit and vegetable production.