InceptionV3-LSTM: A Deep Learning Net for the Intelligent Prediction of Rapeseed Harvest Time

Timely harvest can effectively guarantee the yield and quality of rapeseed. In order to change artificial experience model in monitoring rapeseed period, an intelligent prediction method period based on deep learning network was proposed. Three varieties field were divided into 15 plots, mobile phones used capture images silique stalk manually measure yield. The daily three grades more than 90%, 70–90%, less 70%, according proportion maximum varieties. high-dimensional features canopy extracted using CNN networks HSV space that significantly related maturity rapeseed, seven color stalks screened random forests color-spaces RGB/HSV/YCbCr form a canopy-stalk joint feature as input subsequent classifier. Considering ripening process is continuous time series, LSTM establish classification model. experimental results showed Inception v3 five has highest accuracy. recognition rate 91% when only image used, combined reached 96%. This accurately predict level mature stage by phone take image, it expected become tool for production.