基于改进LSTM的蘑菇生长状态时空预测算法  

Spatiotemporal Prediction Algorithm for Mushroom Growth Status Based on Improved LSTM

作　　者：杨淑珍[1] 黄杰[1] 苑进[2] YANG Shuzhen;HUANG Jie;YUAN Jin(School of Intelligent Manufacturing and Control Engineering,Shanghai Polytechnic University,Shanghai 201209,China;College of Mechanical and Electronic Engineering,Shandong Agricultural University,Taian 271018,China)

机构地区：[1]上海第二工业大学智能制造与控制工程学院,上海201209 [2]山东农业大学机械与电子工程学院,泰安271018

出　　处：《农业机械学报》2024年第3期221-230,共10页Transactions of the Chinese Society for Agricultural Machinery

基　　金：上海市科技计划项目(21N21900600);上海市科技兴农项目(2019-02-08-00-10-F01123);山东省重点研发计划(重大科技创新工程)项目(2022CXGC010609)。

摘　　要：密集蘑菇簇会严重影响蘑菇质量和自动采摘成功率。为避免形成超密集蘑菇簇,提出一种蘑菇生长状态时空预测算法,对蘑菇生长状态进行预测以指导提前疏蕾。该算法采用编码器-预测器框架,将历史序列图像转换为3D张量序列作为模型的输入;编码器网络中将卷积和长短时记忆(Long short term memory, LSTM)网络融合实现对蘑菇生长的时空相关性特征的提取;在预测网络中加入扩散模型以解决预测图像的模糊问题;此外,在损失函数中增加了蘑菇面积差异损失函数来进一步减小预测蘑菇与实际蘑菇的形状和位置偏差。实验结果表明,本文算法峰值信噪比可达35.611 dB、多层级结构相似性为0.927、蘑菇预测准确性高达0.93,有效提高了蘑菇生长状态图像预测质量和精度,为食用菌生长预测提供了一种新思路。Dense mushroom clusters can significantly impact mushroom quality and the success rate of automated harvesting.To address this issue,a spatiotemporal prediction algorithm for mushroom growth status based on historical time series growth images was proposed,which can facilitate early bud thinning to prevent the formation of dense mushroom clusters.The algorithm employed a sequence-to-sequence structure,comprising an encoder and a predictor.In the input,historical image sequences were transformed into 3D tensor sequences and sent to encoder.Within the encoder network,a three-layer long short term memory(LSTM)model was utilized.Here,convolution was fused into LSTM cell to extract spatiotemporal correlation features of mushroom growth.Meanwhile,a diffusion model was introduced into the predictor to address the blurriness issue in predicting images.Furthermore,a mushroom area difference loss function was designed and incorporated into the loss function to further reduce the shape and positional deviations between the predicted and actual mushrooms.The experimental results indicated that the proposed spatiotemporal prediction algorithm for mushroom growth status achieved a peak signal-to-noise ratio of 35.611dB,a multiscale structure similarity of 0.927,and a high mushroom mean intersection over union of 0.93,which represented improvements of 36%,33%and 24%,respectively,over that of the ConvLSTM(Converlution LSTM)spatiotemporal prediction algorithm.This showed the proposed algorithm can effectively enhance the quality and accuracy of mushroom growth status image prediction,providing a approach for precise forecasting of edible mushroom growth.

关 键 词：蘑菇 生长状态预测 长短时记忆网络 扩散模型 面积差异损失函数 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] S126[自动化与计算机技术—控制科学与工程]