Farmland seedling ridge identification method and system based on semanteme and instance segmentation

Abstract

The invention discloses a farmland seedling ridge identification method and system based on semanteme and instance segmentation, and the method comprises the steps: collecting a current farmland image, coding, carrying out instance segmentation and semantic segmentation, clustering, carrying out fitting of feature points, drawing to obtain a seedling line and a furrow line, and the like, whereinthe system comprises a function module for achieving all above steps. According to the invention, seedling ridge detection is carried out by using an end-to-end visual image segmentation training method. The required network scale is reduced to a certain extent, the operation speed is effectively improved. Meanwhile, the embedded vector easy to cluster is obtained through the deep network, then the seedling line and the furrow line are efficiently recognized through the simple clustering algorithm and the curve fitting algorithm, and big data samples can be used for continuously supplementingand improving the accuracy and robustness of the detection method.

Description

technical field [0001] The invention relates to the technical field of intelligent agricultural machinery control, in particular to a method and system for identifying farmland seedling ridges based on semantic and instance segmentation. Background technique [0002] In the process of mechanized agricultural production, accurate seedling ridge identification is the basis of precision agriculture, which can provide space constraints for subsequent precise fertilization, precise weeding, and precise control of intelligent agricultural machinery, so as to achieve the beneficial effects of saving fertilizers and pesticides and increasing yield per mu. Effectively reduce labor intensity, reduce environmental pollution, reduce production costs, improve resource utilization, and increase the output and quality of agricultural products. The current seedling ridge recognition is mostly based on vehicle-mounted visual sensing equipment. The main process is as follows: firstly, the art...

AI Technical Summary

Problems solved by technology

However, the threshold value obtained by repeated adjustments is usually difficult to adapt to all lighting and weather conditions. For example, the lighting at noon on a sunny day is completely different from that on a cloudy evening, resulting in a large difference in the pixel value and super green value of the seedlings in the image. A fixed threshold will inevitably lead to detection failure in one or more scenarios, which will affect the robustness of the method

At the same time, after the traditional method extracts the feature points of the seedling belt, it often needs to distinguish each seedling belt for curve fitting, and distinguish all the corresponding feature points of each seedling belt from the feature points with close distance and similar characteristics. more difficult

Images