当前位置:网站首页>S-RPN: Sampling-balanced region proposal network for small crop pest detection
S-RPN: Sampling-balanced region proposal network for small crop pest detection
2022-07-28 01:18:00 【Romance of cherry blossoms】
abstract
Effective pest control is a key factor in the field of agricultural food safety . therefore , Automatic monitoring and accurate identification of crop pests have high practical value in the process of agricultural planting . these years , With the development of methods based on deep learning , The identification and detection results of pests have improved rapidly . Although these methods are promising , However, the efficiency and accuracy in detecting very small-scale crop pests are still limited , Thus reducing their effectiveness . The main reason is , Current methods based on deep learning may not be able to extract enough detailed appearance features for small pest objects in the image , This makes it difficult to train the classifier to detect and distinguish small pests in the background or similar objects . In order to solve the problem of identification and detection of small pests , In this paper , We seek to rebuild the current regional advice network , And perform more details on different scales , To facilitate the detection of small pests . Inspired by the visual attention system , We first introduce the attention mechanism into the remaining Networks , To get a richer pest characteristic appearance , In particular, the detailed characteristics of small object pests ; then , In order to make the regional recommendation network (RPN) Get higher quality goal suggestions , Easy to detect , A sampling balanced regional proposal generation network is proposed , To improve the accuracy of pest detection . Besides , We design a new adaptive region of interest (RoI) Choose a method to learn features from different levels of the feature pyramid . In what is proposed AgriPest21 Many experiments have been carried out on the data set , Our approach can achieve 89.0% The average recall rate ,mAP achieve 78.7%, Superior to other state-of-the-art methods , Include SSD, RetinaNet, Free-Anchor, PISA, Grid RCNN, and Cascade RCNN
1. Introduction
Agricultural development is limited by various factors , Especially crop pests , It has a negative impact on agricultural production . crop
边栏推荐
- 如果某个表有近千万数据,CRUD比较慢,如何优化
- Postman 的使用
- Learning note 12: Eratosthenes screening method to find prime numbers (within 100) and magic square matrix
- C语言程序设计 | offsetof宏的讲解及其模拟实现
- Use of swarm task task
- mysql-JPA对数据库中JSON类型数据的支持
- Storage of deep planing data in memory
- Six relationships of UML class diagram, the best way to learn and understand
- Code random notes_ Hash_ 1002 find common characters
- Un7.13: how to add, delete, modify and query in vs Code?
猜你喜欢

Demo: the test interface receives duplicate data and creates documents in a short time

Gossip: an initially perfect FS is as impractical as the first version of the program requiring no bugs

Unknown database ‘xxxxx‘

Go 语言变量

DEMO:测试接口短时间内接收重复数据创建单据

The cooperation between starfish OS and metabell is just the beginning

Examples of application of JMeter in performance testing

Analysis and recurrence of network security vulnerabilities
![[300 opencv routines] 241. Scale invariant feature transformation (SIFT)](/img/7a/a764c779c3162920c832325f89f340.png)
[300 opencv routines] 241. Scale invariant feature transformation (SIFT)

Detailed explanation of retinanet network structure
随机推荐
BSP视频教程第21期:轻松一键实现串口DMA不定长收发,支持裸机和RTOS,含MDK和IAR两种玩法,比STM32CubeMX还方便(2022-07-24)
Matlab 绘制 - 点和向量:向量加减的方法和源码
Redefine analysis - release of eventbridge real-time event analysis platform
数据库故障容错之系统时钟故障
0-1 knapsack problem
Monitor mouse sideslip (adapt to mobile terminal)
Redis sentinel mode
EWM收货ECC交货单校验逻辑问题
Jointly create a new chapter in cultural tourism | xinqidian signs a strategic cooperation agreement with Guohua cultural tourism
Behind the war of seizing power in arm China: "two years of independence, it is still difficult to" independent "?
Detailed explanation of swoole memory table
Self use drawing bed building tutorial
110. In depth introduction to sap ui5 fileuploader control - why do you need a hidden iframe
4月全球智能手机出货同比下滑41%,华为首次超三星成全球第一
uni-app进阶之样式框架/生产环境
System clock failure of database fault tolerance
Huawei responded to the US blockade of the supply chain: they still have to pay for 5g patents
Insider of container network hard core technology (7) sailing on the sea depends on the helmsman
[original] [crawler series] briefly get the most popular topics, related topics and description information
Circular structure of shell system learning