Governing Body:South China Agricultural University
Organizers:South China Agricultural University
Chief Editor：Liu Yahong
Contact Address: No.483 Wushan Rd, Tianhe, Guangzhou 510642, Guangdong Province, China
Editing and publishing：Editorial Office of Journal of South China Agricultural University
Service Tel: (020)85280069/38746672
Domestic Distribution:Editorial Office of Journal of South China Agricultural University
Issued Abroad:China International Book Trading Corporation
ISSN : 1001-411X
CN : 44-1110/S
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202309045
The revitalization of seed industry is of great strategic significance to ensure national food security. For promoting the revitalization of Chinese seed industry to ensure food security, we must adhere to the view of great territory and greater food approach (GFA), solve the “strangle hold” problem of seed industry under the dynamic development pattern, pay attention to the necessary matching conditions to revitalize the seed industry considering the coordinated development, and expand the development space to revitalize the seed industry under the perspective of international competition and cooperation. This article revealed the fundamental, critical and decisive roles of seed industry revitalization from the perspective of the priority of factors affecting food security. The development process of Chinese seed industry was reviewed and the relationship between food security and seed industry revitalization was clarified. Under the new pattern of food security, the connotation and challenge of seed industry revitalization were analyzed under the GFA. Some countermeasures and suggestions were put forward for the healthy and orderly development of Chinese seed industry revitalization in the future.
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202305060
Mining rice yield potential is one of the most important directions of rice research. Ratoon rice is a kind of rice that utilizes the axillary buds that survive on the rice pile after the first season of harvesting, the buds germinate into tillers under suitable light, temperature, nutrients and other conditions, and then proceed to heading and maturity. Growing ratoon rice saves a lot of manpower and material resources while ensuring that the rice yield is basically the same as that of double-cropping rice, and it is of great significance for stabilizing national food security. In this paper, the reported QTLs related to rice retooning ability were reviewed, and we found that the contribution rates of the reported QTLs related to rice retooning ability were low, and the related research was not in depth and not closely related to production. The present situation of evaluation criteria for rice retooning ability was summarized. We found that the SPAD attenuation index of the first season rice in the early stage, the sprouting rate of ratoon rice in the middle stage, and the daily yield and heat utilization rate in the late harvest stage were important indicators for evaluating the retooning abilities of rice varieties, which could be used to screen rice varieties with strong retooning abilities. The effects of stump height on yield and quality of ratoon rice were summarized. Finally, we proposed to focus on important rice traits such as high quality, high yield, disease resistance, and other traits related to strong retooning ability, to mine new genes and their excellent natural variation by multi-omics analysis, to develop functional molecular markers, and to combine genome-wide association study with molecular module breeding technology to carry out research on accurate identification of rice germplasm resources with strong retooning abilities, new gene exploration, germplasm innovation and new variety breeding.
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202307001
To introduce the development process and achievements of rice genetic transformation methods, and provide references for the research and application of rice genetic transformation system. Based on the two methods of bio-mediated transformation and abiotic-mediated transformation, the first report and important progress of various transformation methods were introduced, with the prospect at last. Among the bio-mediated transformation methods, Agrobacterium-mediated transformation method is applied by infecting rice embryo, panicle, callus tissue and stem tip. The transformation system of seed embryo and its induced callus as materials is mature, and the transformation method using panicle and stem tip is simple and convenient, with short transformation and regeneration cycle. In addition, preliminary studies have tried to transform rice with Sinorhizobium, Rhizobium and Ensifer adhaerens. Among the abiotic-mediated transformation methods, physical method transformation methods (particle bombardment, electroporation, pollen-tube pathway and microinjection) are more traditional transformation methods. The particle bombardment method is relatively mature, and the pollen-tube pathway method has made great progress in breeding. Among the medium-mediated transformation methods, the application of nanomaterials is gradually becoming a research hotspot. The development of rice genetic transformation can be initiated by selecting transformation materials and optimizing transformation vectors. Concurrently, integrating the transformation system with techniques like DNA-free and haploid induction can enhance both transformation efficiency and safety, while also reducing the duration of the transformation and regeneration process.
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202306065
Rice (Oryza sativa L.) is both a major staple food and a model crop plant for monocot studies. Facing the increasingly severe environmental and population problems, breeding varieties with high yield, high quality, and wide adaptability is the efficient way to solve the food security problems. With the rapid development of multi-omics technology, large volumes of data related to rice genetics and breeding have been accumulated. Bioinformatics databases and online analysis tools are developed to store, integrate, visualize, and share these datasets. In addition, some databases possess built-in tools for further mining and using datasets to provide data support for decision-making in breeding. In this review, we systematically sort out rice bioinformatics databases and online analysis tools developed in the past two decades. Subsequently, we classified and summarized these resources based on their built-in datasets and features. Finally, the problems and deficiencies of the existing rice bioinformatics resources were discussed, and the development direction of bioinformatics resources in the era of big data and artificial intelligence was prospected.
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202306011
The key technologies for low-carbon and high-yield of double-cropping rice in South China (DCRSC) and their integrated application were introduced from the aspects of low-carbon variety selection, water-saving and emission reduction irrigation, and fertilizer reduction and efficiency enhancement, etc. The main progress, problem and future focus on the development and extension of low-carbon and high-yield technology for DCRSC were summarized. The low-carbon and high-yield cultivation technology for DCRSC could synergize water-saving and fertilizer-saving, emission reduction and pollution control, high yield and income, realized the coordination of low-carbon and high-yield, therefore it would have a good application prospect. There are main problems for this technology including limited available varieties, undiversified planting modes, and difficult promotion at present. While increasing policy support, further efforts are needed to enhance the improvement and extension of the new technology, including variety screening, cropping pattern optimization and basic research, so as to ensure national food security and achieve carbon peak and carbon neutrality goals.
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202306069
Objective To reveal the variation characteristics of female semi-sterility in rice induced by space flight. Method The morphologic, cytological, physiological and genetic characteristics of a female semi-sterile mutant rs(s) obtained in conventional indica rice ‘Qiu B’ through space flight mutagenesis were studied. The setting rates of the mutant and its wild type, as well as their reciprocal cross F1 and F2 populations, were investigated. Cytological observations of pollen and embryo sac fertility of the mutant and the wild type were conducted. The content changes of auxin (IAA), cytokinin (TZR), abscisic acid (ABA) and gibberellin (GA3) in young ears of parents were determined. Result The mutant rs(s) was female semi-sterile and it had normal stamen development, indicating the semi-sterility might be controlled by a single recessive gene. During the development, the ovary was shrift and thin, the elongation of the stigma was not obvious, and the pistil was abnormal in the mitotic stage of embryo sac development, which leaded to embryo sac abortion. The content of IAA increased sharply at the 6th stage of young panicle differentiation. Conclusion The mutant rs(s) is a new female semi-sterile mutant. The auxin content is up-regulated in the mitotic stage, which may be the main cause of pistil development malformation and female semi-sterility.
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202306067
Objective To find out the stable QTLs controlling rice seedling height, analyze their epistatic effects, and provide QTL and theoretical references for molecular breeding of rice seedling height.Method The single segment substitution lines (SSSLs) with IR65598-112-2 as donor and ‘Huajingxian 74’ as receptor were used as materials. The difference of seedling height between SSSL and ‘Huajingxian 74’ was measured, and the QTLs of seedling height were mapped. The QTL interval was narrowed by substitution mapping, and the epistatic effects of two seedling height QTLs were also analyzed. Result Two adjacent QTLs (qSH3-1 and qSH3-2) for seedling height were mapped on the long arm of chromosome 3, which were located in the intervals of 32.59-33.08 and 33.16-34.81 Mb, with the lengths of 0.49 and 1.65 Mb, respectively. The additive effects were -0.86 and -1.09 cm, respectively. The phenotypic contribution rate of additive effects were -4.14% and -5.15%, respectively. However, there was no significant difference of seedling height between SSSL harboring these two QTLs and ‘Huajingxian 74’.Conclusion Two QTLs for seedling height were identified, and there may be significant epistasis effects between the two QTLs.
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202306061
Objective To deeply explore new genes related to panicle length and provide a basis for the study of genetic mechanism of panicle length regulation and molecular breeding in rice.Method Two superior parents, ‘ZP37’ and ‘R8605’, as well as 208 recombinant inbred lines (RILs) derived from the cross of ZP37/R8605 were used as a mapping population to locate quantitative trait loci (QTLs) for panicle length in three different environments through the high-density linkage map of whole genome resequencing, and to analyze their pyramiding effects.Result A total of 11 QTLs for panicle length were detected on chromosomes 3, 4, 7, 8, 9 and 12, with the logs of odds (LODs) ranging from 3.07 to 12.87 and contribution rates ranging from 2.17% to 10.94%, seven of the QTLs were new loci, and the remaining four QTLs overlapped or were close to the reported panicle length genes and QTLs. Among them, four stable QTLs were detected repeatedly in two different environments, and by analyzing the lines that pyramiding different numbers of panicle length QTLs, the results showed that the panicle length QTLs showed an additive effect, and the increase in the number of QTLs significantly increased the panicle length of rice. Conclusion The results of this study provide a solid foundation for cloning and functional analysis of rice panicle length QTLs, as well as a theoretical basis and genetic resources for high-yield rice breeding.
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202308015
Objective Long non-coding RNA (lncRNA) is longer than 200 nt and generally does not code proteins. To explore the effect of lncRNA on the expression of neighbouring SAUR genes under low temperature, and provide a theoretical basis for studying the ability of rice seeds to germinate at low temperature. Method lncRNA SVR was investigated in response to cold stress in previous research of National Engineering Research Center of Plant Space Breeding, South China Agricultural University. The secondary structure of lncRNA SVR was analysed by bioinformatics, and the cold stress motifs within lncRNA SVR was searched. The expression characteristics of lncRNA SVR and SAUR genes were analysed by qRT-PCR.Result In this study, highly similar cold stress response motifs were found in the sequence of lncRNA SVR and at the stem-loop junction. Expression characterization showed that cold stress during seed germination continuously decreased the expression of lncRNA SVR, and the expression of some neighbouring SAUR genes was significantly higher under cold stress than that at room temperature, indicating that the neighbouring SAUR genes could respond to cold stress to some extent as well as lncRNA SVR. Expression correlation analysis showed that the expression relationship between lncRNA SVR and these SAUR genes in low temperature germination were all negatively correlated, and the expression of lncRNA SVR and OsSAUR55 showed a significant negative correlation. Further analysis of SAUR gene expression in low temperature seed germination in the lncRNA SVR knockout lines showed that the decrease of lncRNA SVR in the knockout lines was lower than that in the wild type, and the increase in expression of OsSAUR41, OsSAUR53, OsSAUR54 and OsSAUR55 was significantly lower than that in the wild type.Conclusion lncRNA SVR is able to negatively regulate the expression of OsSAUR55, further respond to low temperature stress.
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202306020
Objective To identify the variation in heat stress resistance among seedlings of four different indica rice cultivars grown in South China, and provide theoretical reference and technical support for the breeding and promotion of rice varieties. Method We performed whole-genome re-sequencing, phenotype identification, and transcriptional level analysis to comprehensively evaluate the heat stress resistance for seedlings of high-yield conventional rice cultivar ‘Nan Xiu Mei Zhan’ (NXMZ), hybrid rice restorer cultivar ‘R5518’, and aromatic rice cultivars ‘Jiu Li Xiang’ (JLX) and ‘Nan Jing Xiang Zhan’ (NJXZ) in South China.Result The ‘NJXZ’ was sensitive to heat stress. The ‘R5518’ showed medium resistance to heat stress. The resistances to heat stress of ‘JLX’ and ‘NXMZ’at seedling stage were relatively high in comparison to other two rice cultivars. We compared the haplotypes of heat resistance related genes, the relative expression levels and phenotypes, and found that many SNPs appeared in OsTT1 from‘JLX’, while the haplotypes of four cultivars with other heat-related QTLs remained relatively consistent, suggesting that the OsTT1 might contribute to partial heat stress resistance in ‘JLX’. The gene expression patterns in OsHSF7, OsHSP71.1 and OsHTS1 were consistent with the evaluation in heat stress resistance of rice cultivars, indicating that these three genes might associate with regulation in heat stress resistance of four indica rice cultivars.Conclusion The gene expression difference, gene shift and transcript error in certain genes result in variations in heat stress resistance of different indica rice cultivars at seedling stage. These results can provide new ideas for genome-wide selection for heat tolerance breeding in rice.
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202308020
Objective The goal was to identify the aerobic adaptation of early seedlings of upland rice germplasm resources, screen out the superior germplasm resources with the abilities of seed germination under low water stress and seedling emergence from deep soil, and provide genetic materials and valuable guidance for breeding new viarieties with excellent aerobic adaptation.Method Under the conditions of 150 g/L polyethylene glycol (PEG6000) mimicking low water stress and 8 cm soil deep-seeding, 24 lowland rice (including ‘Huanghuazhan’, ‘longjing39’ and ‘jingyue1’) and 24 improved upland rice (including ‘Luying46’,‘IRAT318’ and ‘IRAT11’) varieties were used to evaluate the aerobic adaptability indicators. In addition, these indicators were assessed for 246 upland rice landraces. The principal component analysis, membership function and comprehensive evaluation value of aerobic adaptation were used to evaluate these germplasms, and they were classified by cluster analysis. Result The relative germination potential, germination rate, root length, bud length and root number between lowland and upland rice varieties under low water stress were significantly different, and could be used as indicators for identifying aerobic adaptation at the early seedling stage. Under deep-seeding condition, root length, shoot length, mesocotyl length and coleoptile length were mainly responsible for seedling emergence. According to their differences in aerobic adaptability, the germplasms could be divided into five types, including highly strong type (type Ⅰ), strong type (type Ⅱ), intermediate type (type Ⅲ), weak type (type Ⅳ) and extremely weak type (type Ⅴ). Finally, seven varieties with elite aerobic adaptability were obtained.Conclusion This study identified the indicators of aerobic adaptation and comprehensively assessed the upland rice landraces in Yunan, which could provide a basis for fostering dry direct-seeding rice varieties.
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202306066
Objective To explore the information of rice lysophosphatidic acid acyltransferases (LPAT) gene family, and analyze the bioinformatics characteristics and its role in rice grain oil synthesis. Method Through bioinformatics analysis, the gene structure, phylogenetic tree, tissue expression profiles, hormone and stress expression profiles of OsLPAT gene family were analyzed. The roles of OsLPAT family members in rice grain oil synthesis were analyzed by comparative metabolomics and transcriptome methods. Result The rice genome contain five LPAT genes, named OsLPAT1-OsLPAT5. Except OsLPAT1, other members all contain 4-6 exons, and each member contains the acyltransferase C-terminal (PF01553) domain. The evolutionary analysis showed that the LPAT genes were relatively conservative in monocotyledon. OsLPAT2 was sensitive to various stress conditions and hormones. The expressions of OsLPAT3 and OsLPAT5 were strongly induced by abscisic acid (ABA), while OLPAT4 was specifically induced by osmotic stress. Comparative metabolomics and transcriptome analysis showed that each member of the OsLPAT family played roles in different stages of grain development, thereby promoting the synthesis of oil triacylglycerol (TAG). Conclusion There are five LPAT genes in rice genome, whose expression are induced by different hormones and may play roles in the hormone response process of rice. The members of this family are involved in the synthesis of phosphatidic acid (PA), the oil precursor in grains at different development stages, thus positively regulating the oil synthesis in rice grains.
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202304017
Objective The study aims to analyze the dynamic changes of panicles per plant, leaf area index (LAI), leaf SPAD, and light transmittance rate (LTR) in late-season indica hybrid rice, further clarify their interrelationships and their effects on yield and yield-related traits, and provide a theoretical guidance for hybrid rice breeding and production practice.Method Five three-line sterile lines and six restorer lines, which are widely used in South China, were used as hybrid combination parents, and a random block experiment of 27 hybrid combinations was conducted in Guangzhou in the late season of 2021. The dynamic changes of photosynthetic parameters, the effects of photosynthetic parameters at different growth and development stages on yield and yield-related traits, and their dynamic correlation were studied. Result The number of hybrid tillers increased linearly after transplantation, peaked on 25 d after transplantation, and entered a stable period at the initial heading stage (60 d after transplanting). The hybrid LAI rapidly increased after transplanting, reaching its highest value at the late stage of spike differentiation (50 d after transplanting), and then entered a decline period. The leaf SPAD of hybrids gradually decreased after transplantation, and there was no significant difference at the early growth and development stage, while there was significant (P < 0.05) or extremely significant (P < 0.01) differences between the SPAD at the grain-filling stage. The LTR of hybrid populations decreased gradually with the development process. The correlation analysis showed that there was a highly significant positive correlation between the number of tillers per plant and hybrid yield at peak tillering stage (10−20 d after transplanting), initial heading stage (60 d after transplanting), and grain-filling stage (76 d after transplanting). The yield increase effect was mainly achieved by increasing the number of filled grains per plant. The excessive number of tillers during peak tillering stage to the late stage of spike differentiation (25−50 d after transplanting) increased the ineffective tillers, resulted in a decrease in hybrid seed setting rate and a significant decrease in yield. The hybrid LAI value at tilering stage (20 d after transplanting) and initial heading stage (60 d after transplanting) showed a significant or extremely significant positive correlation with yield, with the correlation coefficients of 0.296 and 0.255, respectively. The yield increase effect were mainly achieved by increasing the number of filled grains per plant and the weight of 1000-grain. The LAI value at grain-filling stage (76 d after transplanting) showed a highly significant negative correlation with yield with the correlation coefficient of -0.312. The SPAD of hybrids at the early stage of growth (15−50 d after transplanting) had a significant or extremely significant increase in yield, while the SPAD value in the later stage (76−90 d after transplanting) caused a very significant decrease in yield. The LTR of hybrid populations showed extremely significant negative correlation with their yield, the correlation coefficients between LTR and yield at tillering stage (20 d after transplanting), and early stage of spike differentiation (38 d after transplanting) were -0.282 and -0.384, respectively.Conclusion The combinations derived from ‘Tianfeng-A’, ‘Wufeng-A’, ‘R998’ and ‘R308’ have strong early tillering ability, a large number of stem tillers, a large LAI, and good early growth and rapid development. The combinations of ‘Yangtai-A’ and ‘R998’ show a lower LTR in the early growing stage and a higher LTR in the later stage, which is much beneficial for plant photosynthesis and yield improvement. The yield of hybrids is affected by photosynthetic parameters at different stages of growth and development, through influencing the different yield traits. By fitting the regression equation between photosynthetic parameters and hybrid yield, the yield of hybrids can be well predicted at early growing stage.
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202306068
Objective To find the characteristics of responses of eating quality related traits to nitrogen fertilizer application rates and planting seasons, identify rice varieties (lines) with different responding types, screen out fine quality ones which are insensitive to nitrogen application amount, and provide theoretical and technical supports for developing new rice varieties (lines) with fine and stable quality and establishing high-yield and high-quality cultivation techniques. Method Totally seventeen rice varieties or lines were involved, including high-quality and high-yield conventional and hybird rice varieties, and their parental lines from South China, and four treatments of nitrogen fertilzer application rates (including 0, 90,135 and 180 kg/hm2) were conducted in the early and late cropping seasons, respectively, and the response of cooking and eating quality was observed. Result 1)Except that the alkali spreading value (ASV) had fewer response, amylose content (AC), protein content (PC) and taste value (TV) all had the obvious response to nitrogen fertilzer application rate, suggesting that the nitrogen fertilizer had significant affect on these quality traits. The PC increased with the increase of nitrogen application, whereas TV decreased with the increase of nitrogen application. There exsisted the interaction effect of nitrogen application level×variety (line) for Gel consistence (GC) in both early and late seasons, and for AC、PC and TV in the late season, suggesting that the effect of nitrogen application rate on AC, GC, PC and TV varied with varieties. 2)In the late season (LS), the rice varieties (lines) treated with no nitrogen fertilizer (N0) usually had the highest taste value, while in the early season (ES), the rice varieties (lines) treated with 90 kg/hm2 of nitrogen (N1) generally had the highest taste value, suggesting that a small amount of nitrogen fertilizer should be applied in ES, whereas as far as possible no or very few amount of nitrogen fertilizer applied in LS may be more conducive to improving the taste of rice. 3)Different high-quality rice varieties (lines) had different sensitivity to nitrogen application in rice taste, there exist great differences among varieties. 4)The GC, AC, PC and TV of hybrid F1 were mostly between their parents. However, their TV tended to the high-value parent, and exhibited partial dominant or overdominant. Conclusion To develop conventional rice and hybrid rice with good taste, it is necessary to use materials with medium and low PC and higher TV as parents for improving the breeding efficiency of new rice varieties. In the production of high quality rice, selecting high quality rice varieties with good taste and lower sensitivity to nitrogen fertilizer, and matching corresponding fertilization technology program according to the production season, is crucial to the development of high-taste and high-quality rice industry.
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202209015
Objective In order to improve the grading detection accuracy and speed of hybrid rice seed vigor. Method A rapid grading detection method for hybrid rice bud seeds named YOLOv5-I model, which was an improved model based on YOLOv5, was proposed. The feature extraction ability of the target channel of YOLOv5-I model was improved by introducing the SE (Squeeze-and-excitation) attention mechanism module, and a CIoU loss function strategy was adopted to improve the convergence speed of this model. Result The YOLOv5-I algorithm effectively achieved the rapid grading detection of hybrid rice bud seeds, with high detection accuracy and speed. In the test set, the average accuracy of the YOLOv5-I model was 97.52%, the average detection time of each image was 3.745 ms, and the memory space occupied by the YOLOv5-I model was small with 13.7 MB. The detection accuracy and speed of YOLOv5-I model was better than those of YOLOv5s, Faster-RCNN, YOLOv4 and SSD models. Conclusion The YOLOv5-I algorithm is better than existing algorithms, improves detection accuracy and speed, and can meet the practical requirement for grading detection of hybrid rice bud seeds.
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202208022
Objective The measurement accuracy and efficiency of traditional seed investigation methods could not meet the needs of modern rice breeding research. A synchronous collection device of rice grain image and quality information was designed to automatically extract rice seed investigation parameters.Method The image of grain region was automatically extracted by mask method, and the total number of rice grains was obtained according to the law between rice projection area and rice quantity. Empty grains were identified according to the difference of the hull contour between empty grains and full grains. Based on the mean value calibration method of corner spacing, the grain length and width were obtained by combining the minimum circumscribed rectangle method of contour, and the grain perimeter was obtained by combining the chain code method. The square area mean calibration method and pixel accumulation method were used to obtain the grain area. The effects of camera height, grain quantity, grain type and regular graph type on the extraction accuracy of grain character parameters were analyzed.Result The camera height had a obvious impact on the measurement accuracies of total number, empty number, length and width of rice grains, the grain type had a obvious impact on the measurement accuracy of width, and the regular graph type had a obvious impact on the measurement accuracies of grain area and perimeter. The determination coefficients (R2) of total grain number, empty grain number, grain length, grain width, grain perimeter and grain area measured by the proposed method were 0.99830, 0.98780, 0.99610, 0.78290, 0.99510 and 0.99998 respectively, the average measurement accuracies were 99.47%, 87.17%, 96.55%, 96.36%, 98.00% and 95.86% respectively, and the measurement efficiency was 16.52 grains per second. Conclusion The automatic extraction method of rice seed investigation parameters used in this paper is feasible, and can provide technical references for the development of automatic seed investigation machine.
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202208020
Objective The yield of rice is related to food security of all mankind. How to effectively prevent and detect rice diseases and pests is an important topic in the field of smart agriculture. Deep learning has become the preferred method for identifying rice diseases and pests due to its excellent performance in self-learning image features. However, in natural environments, the dataset is relatively small and susceptible to complex backgrounds, resulting in overfitting and difficulty in extracting subtle features during training. This study aims to address the aforementioned issues. Method We proposed a rice disease and pest identification model with multi-scale dual branch structure based on improved ResNet (MSDB ResNet). On the basis of the ResNet model, ConvNeXt residual blocks were introduced to optimize the calculation proportion of residual blocks, construct a dual branch structure, and extract disease features of different sizes from the input disease image by adjusting the convolution kernel size of each branch. In response to issues such as complex real world environments, small datasets, and overfitting, a total of 5932 rice pest and disease images captured from natural environments was utilized. Using data preprocessing methods such as random brightness and motion blur, as well as data augmentation methods such as mirroring, cropping, and scaling, the dataset was expanded to 20000 pictures. The MSDB-ResNet model was trained to identify four common rice diseases. Result MSDB-ResNet had good recognition performance on rice disease and pest datasets, with a recognition accuracy of 99.10%, which was 2.42 percentage points higher than the original ResNet model and obviously superior to classic networks such as AlexNet, VGG, DenseNet, ResNet, etc. This model had good generalization ability and strong robustness. Conclusion The MSDB ResNet model is feasible and progressiveness in the identification of rice diseases and pests, which provides a reference for the identification of rice diseases and pests under complex background.
2023,44(6) DOI: 10.7671/j.issn.1001-411X.202208051
Objective To solve the problems that the current pest identification method has many parameters, a large amount of calculation and is difficult to deploy embedded devices at the edge, so as to realize accurate identification of crop pests and diseases, and improve crop yield and quality.Method A lightweight convolutional neural network called multi-head attention to convolutional neural network (M2CNet) was proposed. M2CNet adopted hierarchical pyramid structure. Firstly, a local capture block was constructed by combining depth separable residual and cyclic fully connected residual to capture short-range information. Secondly, a lightweight global capture block was constructed by combining global subsampling attention and lightweight feedforward network to capture long-distance information. Three variants, namely M2CNet-S, M2CNet-B, and M2CNet-L, were proposed by M2CNet to meet different edge deployment requirements.Result M2CNet-S/B/L had parameter sizes of 1.8M, 3.5M and 5.8M, and floating point operations of 0.23G, 0.39G, and 0.60G, respectively. M2CNet-S/B/L achieved top5 accuracy greater than 99.7% and top1 accuracy greater than 95.9% in PlantVillage disease dataset, and top5 accuracy greater than 88.4% and top1 accuracy greater than 67.0% in IP102 pest dataset, outperforming models of the same level in comparison.Conclusion Effective identification of crop diseases and pests can be achieved by this method, and it provides valuable references for edge engineering deployment.