Small target detection algorithm based on dense deconvolution and specific loss function

Abstract

The invention discloses a small target detection algorithm based on dense deconvolution and a specific loss function, and the algorithm comprises the steps: 1, extracting image features: extracting a multi-layer convolution output result to form a multi-feature map, extracting target regions of interest of different visual fields from the multi-feature map, and carrying out the feature connection of the extracted target regions of interest; 2, using a weighted linear combination of independent loss functions to optimize and generate a comprehensive loss value function, and obtaining an overall loss function through training minimization; 3, carrying out semantic segmentation on the original image, extracting a target segmentation area result, extracting target interested areas of different visual fields from the multi-feature image,determining an anger table area. The whole algorithm is high in accuracy and high in practicability. By optimizing and generating the comprehensive loss value function by using the weighted linear combination of the independent loss functions and obtaining the overall loss function through training minimization, the steps can be simplified, and the stability of target detection can be improved.

Description

technical field [0001] The invention relates to the technical field of deconvolution, in particular to a small target detection algorithm based on dense deconvolution and a specific loss function. Background technique [0002] Deconvolution refers to the process of reconstructing unknown inputs from measured outputs and known inputs; in statistics, statistical decision theory, and economics, a loss function refers to a method that maps an event to an expression that expresses the economic cost associated with its event Or a function on the real numbers of opportunity costs. More generally speaking, in statistics, a loss function is a function that measures the degree of loss and error. [0003] The current detection algorithm for deconvolution and loss function is not accurate enough, and the stability is poor, and the whole model is too complex to be realized. Therefore, the present invention provides a small target detection based on dense deconvolution and specific loss ...

AI Technical Summary

Problems solved by technology

[0003] At present, the detection algorithm for deconvolution and loss function is not accurate enough, and the stability is poor, and the whole model is too complicated to be realized. Therefore, the present invention provides a small target detection based on dense deconvolution and specific loss function algorithm