Improved Faster RCNN hydroponic vegetable seedling state detection method

Abstract

The invention discloses an improved Faster RCNN hydroponic vegetable seedling state detection method, and belongs to the technical field of agricultural cultivation. According to the method, an improved Faster RCNN detection network is provided, aiming at the characteristic that hydroponic vegetable seedling images are tiny and dense, HRNet is adopted as a feature extraction network, information loss in the down-sampling process is reduced, information of small target objects is well reserved, and therefore good semantic information is provided for regression and classification of follow-up candidate boxes. According to the method, the RoI Align and Soft NMS methods are applied to improve the recognition precision and recall rate of the model, so that the detection effect of the model is improved, a good and reliable feature map is obtained, and a good foundation is laid for subsequent candidate box classification and regression; in the modernization process of the facility agriculture, a deep learning algorithm is migrated to the field of seedling images of the facility agriculture, automatic detection of the seedling state of the hydroponic vegetables is achieved, and manual labor is reduced.

Description

technical field [0001] The invention belongs to the technical field of agricultural cultivation, in particular to an improved Faster RCNN hydroponic vegetable seedling state detection method. Background technique [0002] As the saying goes: "The seedlings are strong and half harvested", seedling raising is a labor-intensive, time-consuming and highly technical job. At present, in the process of growing seedlings of hydroponic vegetables, the required cultivation time is long, the labor intensity is high, and the labor cost is high. During the field investigation of the growth process of hydroponic vegetables, it was found that the growth of the seedlings on the sponge was different for the hydroponic vegetables in the seedling stage. Specifically, there were no seedlings growing in some holes, due to the There is a certain emergence rate in germination, and there are cases where the sown seeds do not germinate; in some holes, two seedlings grow. This situation has potentia...

AI Technical Summary

Problems solved by technology

Using the VGG16 feature extraction network to test on the PASCAL VOC 2007 data set achieved an accuracy rate of 73.2%, but Faster R-CNN uses ROI pooling, which makes the network features lose translat

Images