Ship object recognition method based on multilayer convolution neural network

Abstract

The invention discloses a ship object recognition method based on a multilayer convolution neural network, and the method comprises the steps: S1, employing the existing images, parameters and model data for the building of a ship sample library, and continuously increasing the data in a use process through target detection; S2, carrying out the ship target feature training under the frame of a convolution neural network, and forming visible light / infrared and two-dimensional / three-dimensional fusion ship feature knowledge library through the recognition training of the ship sample library, wherein the ship feature knowledge library is used for the classification and recognition of ships; S3, carrying out the data collection of a ship target: carrying out the real-time high-resolution collection of the visible light or infrared video data of the ship target on the sea; S4, carrying out the detection of the ship target on the sea; S5, carrying out the coarse classification of the imagesof the ship target; S6, carrying out the fine classification and recognition of the ship target based on a depth neural network completed through the ship target feature training, and accurately recognizing the type of the recognized ship. The method overcomes a difficulty in recognition of the ship target.

Description

technical field [0001] The invention relates to a target recognition method, in particular to a ship target recognition method based on a multi-layer convolutional neural network. Background technique [0002] my country has a vast coastline, sea areas and rich marine resources. With the continuous development of the economy, the number of ships at sea is increasing, and there is an urgent actual need for ship inspection; and ships and civilian ships in neighboring countries and regions are often illegal. Entering my country's legal waters to engage in activities such as surveying, monitoring, and fishing, so that the country's legal maritime rights and interests cannot be effectively guaranteed, the situation of marine rights protection is becoming more and more complicated, and maritime security is seriously threatened. The detection and identification of detection has very important practical significance. [0003] At present, the ability of image reconnaissance in my coun...

AI Technical Summary

Problems solved by technology

Due to the long distance between satellite data and the ground, the transmission speed is slow, the ship target data cannot be updated in real time, and ships appearing in the designated sea area cannot be found and identified in time; satellite remote sensing cannot clearly capture sea surface images in thunderstorm and cloudy weather, resulting in severe sea surface The accuracy rate of ship target recognition in the environment is reduced

At the same time, various types of military and civilian ships are updated very quickly. The ship recognition image library generated by the general method needs to be upgraded frequently, which increases the maintenance time and cost of the ship, and the degree of intelligence is not high.

Images